MOLECULAR CLASSIFIERS FOR PROSTATE CANCER

Abstract

There is described herein a method of predicting disease progression risk in a subject with prostate cancer, the method comprising: a) providing a sample containing RNA and DNA material from tumour cells; b) determining or measuring values for substantially all of patient features listed for PRONTO-e or PRONTO-m in Table 6, and some or all reference or control features set forth in Table 6; c) comparing said patient features to the reference or control features; and d) computing a prediction score using a classifier that takes said patient feature values as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

Description

FIELD OF THE INVENTION

The invention relates to molecular classifiers and more particularly to classifiers for prostate cancer.

BACKGROUND OF THE INVENTION

Although prostate cancer (CaP) is a leading cause of cancer death, the majority of biopsy-confirmed cases are sufficiently indolent to be safely monitored without definitive treatment [1, 2]. The most powerful biomarker of aggressive prostate cancer has been Gleason Grade, as determined by comprehensive pathologic examination of the surgically removed prostate. Low Gleason grade cancers, defined as Gleason grade 3+3=6 or WHO Grade Group (GG) 1 [3], exhibit negligible risk of metastasis or death [4, 5]. Higher-grade cancers (WHO GG2 to GG5) require definitive treatment. Unlike most cancer types, for which grading schemes prioritize nuclear morphology and mitotic counts, GG for prostate cancer focuses exclusively on glandular architecture. Both benign prostate glands and glands formed by GG1 prostate cancer cells feature a single layer of luminal epithelial cells surrounding a single lumen. All cancer cells occupy similar environments, directly contacting the lumen on apical aspects, with stroma at their base, and other cancer cells on the remaining four sides. This arrangement provides similar access to oxygen and nutrients from surrounding blood vessels. In contrast, higher grade cancers (GG2-GG5) form fused gland-like structures with multiple lumens, or make no lumens at all, reflecting far greater plasticity with respect to cell-cell interactions, differentiation, and metabolism.

The ability to grow in these different arrangements corresponds to the ability to grow as metastatic deposits outside the prostate. Thus, cancer metabolism, epithelial plasticity, and epithelial-stromal interactions are key themes in prostate cancer progression [6-9]. The molecular underpinnings of glandular architecture associated with GG provide direction for the development of diagnostic biomarkers for aggressive prostate cancer.

In the United States, Canada, and Europe, active surveillance (AS) represents a standard of care for GG1 cancers [10-13]. Patients are monitored with prostate-specific antigen (PSA) levels and a series of core biopsies and may receive imaging as an adjunct [10]. While GG based on prostatectomy is highly informative, current methods cannot accurately separate GG1 and GG2 based on needle biopsies, presenting a major dilemma. Due to sampling error in core biopsy and inter-observer variability, biopsy grading inaccurately reflects surgical GG in 36-67% of cases [14-17]. The consequence of these inaccuracies is that men are placed into the wrong risk category. Those who are eligible for AS may receive aggressive surgical interventions (radical prostatectomy) and suffer undue morbidity, due to uncertainty relating to their true risk of harboring aggressive high-grade cancer. Conversely, others fail to receive the treatment they require in time to prevent the spread of incurable metastatic disease.

Inaccurate reporting of GG at biopsy has motivated molecular approaches to improving risk stratification based on a core biopsy sampling of CaP [18]. However, existing molecular classifiers for biopsy GG fail to accurately distinguish between GG1 and GG2 [19, 20].

SUMMARY OF THE INVENTION

In an aspect, there is provided a method of predicting disease progression risk in a subject with prostate cancer, the method comprising: a) providing a sample containing RNA and DNA material from tumour cells; b) determining or measuring values for substantially all of 353 patient features comprising the mRNA and copy number aberration (CNA) features listed for PRONTO-e in Table 6, and some or all reference or control features set forth in Table 6; c) comparing said patient features to the reference or control features; and d) computing a prediction score using a classifier that takes said patient feature values as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

In an aspect, there is provided a method of predicting disease progression risk in a subject with prostate cancer, the method comprising: a) providing a sample containing RNA and DNA material from tumour cells; b) determining or measuring substantially all of 94 patient features comprising the mRNA, CNA, methylation and clinical features listed for PRONTO-m in Table 6, and some or all reference or control features set forth in Table 6; c) comparing said patient features to the reference or control features; and d) computing a prediction score using a classifier that takes said patient feature values as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

In an aspect, there is provided a computer-implemented method of predicting disease progression risk in a patient with prostate cancer, the method comprising: a) receiving, at at least one processor, data reflecting substantially all of the patient features defined in claim 1 or 7 corresponding to the PRONTO-e or PRONTO-m classifiers regarding a prostate cancer tumor, and some or all reference or control features set forth in Table 6; b) constructing, at at least one processor, a patient profile based on the patient features; c) comparing, at the at least one processor, said patient profile to the reference or control; d) computing, at the at least one processor, a prediction score using a classifier that takes said patient profile as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

In an aspect, there is provided a computer program product for use in conjunction with a general-purpose computer having a processor and a memory connected to the processor, the computer program product comprising a computer readable storage medium having a computer mechanism encoded thereon, wherein the computer program mechanism may be loaded into the memory of the computer and cause the computer to carry out the method of any one of claims 13-15.

In an aspect, there is provided a computer readable medium having stored thereon a data structure for storing the computer program product according to claim 16.

In an aspect, there is provided a device for predicting disease progression risk in a patient with prostate cancer, the device comprising: at least one processor; and electronic memory in communication with the at least one processor, the electronic memory storing processor-executable code that, when executed at the at least one processor, causes the at least one processor to: a) receive data reflecting substantially all of the patient features defined in claim 1 or 7 corresponding to PRONTO-e or PRONTO-m classifiers regarding the prostate cancer tumor, and some or all reference or control features set forth in Table 6; b) compare said patient features to the reference or control features; and c) compute, at the at least one processor, a prediction score using a classifier that takes said patient profile as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

BRIEF DESCRIPTION OF FIGURES

These and other features of the preferred embodiments of the invention will become more apparent in the following detailed description in which reference is made to the appended drawings wherein:

FIG. 1. Overview of approach.

(A) Cases were split into training and validation cohorts. Both high-grade and low-grade samples were extracted from each resected tumor (i.e. for each case). (B) 431 genes/loci associated with GG were profiled. (C) A machine learning pipeline was used to develop GG classifiers. First, one or more data types were selected. Second, the relevant data were partitioned for five-fold cross-validation. Third (optional), features without significant univariate association with GG were discarded. Fourth, after selecting a machine learning algorithm, a classifier was trained on four partitions and tested on the 5th partition.

FIG. 2. Performance of the top-25 classifiers from repeated cross-validation.

Each column represents a classifier. The top panel indicates the datasets used by the classifier, the machine learning algorithm used to train it, the sample weighting (i.e. envelope) scheme and the types of training samples used (see Methods). In the AUC panel, each box summarizes the mean AUCs from the 1000 repetitions of cross-validation. In the GG1 and GG2 panels, each box summarizes the mean fractions of correctly classified GG1 and GG2 cases, respectively. The mean statistics were computed as x_mean=(x_low+x_high)/2 where x_low and x_high are the statistics computed from only low- or high-grade samples, respectively. The classifiers are sorted by decreasing AUC. Abbreviations: AUC—area under the curve; BCR—biochemical recurrence; CAPRA—Cancer of the Prostaste Risk Assessment; CN_MLPA—copy number, MLPA platform; CN_NS—copy number, NanoString platform; GG—grade group; MSP—methylation-specific PCR.

FIG. 3. Performance of multimodal classifiers PRONTO-e and PRONTO-m.

(A-C) Multimodal classifiers, i.e. classifiers that use different types of data, outperform single-mode classifiers in cross-validation. The TP rate (A), FP rate (B) and AUC (C) of each classifier were computed from cross-validation repeated 1000 times (boxes summarize the repetitions). In each repetition, each statistic was computed using only the high- or only the low-grade sample from each case. The mean of the high- and low-grade statistics is indicated in the ‘mean’ section. The type of input data used by a given classifier is indicated in the key in (C); CAPRA uses only clinical data. The multimodal classifiers are top-performing classifiers according to cross-validation. (D) Validation performance of the multimodal classifiers. For each case in the validation cohort, one sample was randomly selected and statistics were computing using the representative samples. This process was repeated 1000 times and each point indicates the median across repetitions (i.e. sampling-based AUC), and the lower and upper error bars indicate the first and third quartiles, respectively. (A-C) CNA refers to CNA data from MLPA since PRONTO-e and PRONTO-m only use CNA data from MLPA. (E) Agreement between predicted classes of low- and high-grade samples from the same validation case. (F) Of those cases with agreement, the percentage with a correct prediction. (E-F) Cases with GG1 are separated from patients with GG2. The total numbers of validation cases used to compute each percentage are shown above the bars. Note that the numbers vary for PRONTO-e and PRONTO-m since the classifiers have different data requirements for each sample.

FIG. 4. Molecular features with significant univariate associations with GG (q-value <0.1).

For each significant molecular feature, the left plot shows the median difference in feature values for GG≥2 and GG1 cases. The difference is show for each cohort, where the point indicates the median and the ends of the intersecting line indicate the first and third quartiles, across 1000 random selections of one representative sample per case. The right plot indicates the q-value (i.e. adjusted p) resulting from the combination of the training and validation cohort q-values, representing the significance of the univariate association between the feature and GG (see Methods). The mRNA feature analysis used 332 training and 200 validation cases, and the methylation feature analysis used 318 training and 202 validation cases. For the targeted genes, preferential expression in the epithelial or stromal compartments is indicated [54].

FIG. 5. Computer device for implanting the methods.

A suitable configured computer device, and associated communications networks, devices, software and firmware to provide a platform for enabling one or more embodiments as described herein.

FIG. 6. Overview of the GG classifier design.

The GG classifier would take a patient profile as input, where the profile is potentially comprised of features of different data types (including clinical features, not shown). The classifier is trained with one of several possible machine learning algorithms (see Methods) to predict whether the patient has a pathological GG2 or not. That is, the final classifier output would be yes or no.

FIG. 7. PRONTO-e and PRONTO-m at different operating points.

(A) Validation ROC curves of the PRONTO-e and PRONTO-m classifiers on only the low-grade or only high-grade sample from each case. The prediction score is the numerical output of a classifier, and with an operating point of x, a score>=x predicts pathological GG>=2 whereas a score<x predicts pathological GG1. Curves show the true and false positive rates at different operating points. (B) The prediction score distributions of the PRONTO-e and PRONTO-m classifiers. Boxes indicate the score distributions from the classifiers applied to all samples in the validation cohort, separated by the GG of their source cases. As expected, the scores tend to be higher for samples from higher GG cases, for both classifiers. The red line indicates the chosen operating point of 0.5.

FIG. 8. Similarity between the molecular profiles of the low- and high-grade samples from the same case.

CNA refers to CNA data from MLPA since PRONTO-e and PRONTO-m only use CNA data from MLPA. Abbreviation: methyl.—methylation.

FIG. 9. Potential clinical impact of PRONTO-e.

Hypothetical performance of the PRONTO-e classifier if applied to the diagnostic biopsy of 1000 patients recommended for active surveillance. Given 1000 active surveillance patients and the predicted performance of PRONTO-e, the illustration shows the hypothetical number of true and false positives, true and false negatives, and how these patient subsets would be impacted by their test results. A positive test result would trigger an early biopsy 3 or 6 months after diagnosis, which may result in upgrading and subsequent treatment. A negative test result would instead lead to a biopsy 12 months after diagnosis.

DETAILED DESCRIPTION

In the following description, numerous specific details are set forth to provide a thorough understanding of the invention. However, it is understood that the invention may be practiced without these specific details.

Cancer grade is the most powerful predictor of disease progression in early-stage prostate cancer (CaP). Infra-tumoral heterogeneity and inter-observer variability limit accuracy in diagnostic biopsies, and reduce clinical utility. Using pathologic examination of the prostatectomy as the gold standard, we developed and validated a robust objective biomarker of prostate cancer grade.

Radical prostatectomies were collected from low- and intermediate-risk CaP patients and assigned to either a training (n=333) or validation (n=202) cohort. To integrate intra-tumoral heterogeneity, each case was separately sampled at two locations. We profiled 342 mRNAs enriched for CaP metabolism, stromal signaling, and epithelial plasticity, complemented by 100 copy number aberrations (CNAs) and 14 DNA hypermethylation loci. Over 41,000 candidate classifiers of pathologic Grade Group (1 versus 2) were generated with the training data, subjecting clinical, pathologic and molecular variables to 12 different machine learning algorithms. We selected two classifiers, PRONTO-e and PRONTO-m, for validation by prioritizing classifiers with greater true positive (TP) rates and areas under the receiver-operator curve (AUCs).

The PRONTO-e classifier comprises 353 mRNA and CNA features, while the PRONTO-m classifier comprises 94 mRNA, CNA, methylation and clinical features. The classifiers (PRONTO-e, PRONTO-m) independently validated, with respective true positive rates of 0.802 and 0.810, false positive rates of 0.403 and 0.398, and AUCs of 0.799 and 0.786.

Two multigene classifiers were developed and validated in separate cohorts, each achieved excellent performance by integrating different types of genomic data. Classifier adoption could improve current active surveillance approaches without increasing patient morbidity.

In an aspect, there is provided a method of predicting disease progression risk in a subject with prostate cancer, the method comprising: a) providing a sample containing RNA and DNA material from tumour cells; b) determining or measuring values for substantially all of 353 patient features comprising the mRNA and copy number aberration (CNA) features listed for PRONTO-e in Table 6, and some or all reference or control features set forth in Table 6; c) comparing said patient features to the reference or control features; and d) computing a prediction score using a classifier that takes said patient feature values as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

In some embodiments, substantially all of 353 patient features is all 353 patient features.

As used herein, the term “control” refers to a specific value or dataset that can be used to prognose or classify the value e.g. patient features comprising the mRNA, copy number aberration (CNA) features, or clinical features obtained from the test sample associated with an outcome class. A person skilled in the art will appreciate that the comparison between the test sample and the control will depend on the control used.

The term “low risk” or “low likelihood” as used herein in respect of cancer refers to a statistically significant lower risk of cancer as compared to a general or control population. Correspondingly, “high risk” or “high likelihood” as used herein in respect of cancer refers to a statistically significant higher risk of cancer as compared to a general or control population.

The term “sample” as used herein refers to any fluid, cell or tissue sample from a subject that can be assayed for the DNA or RNA materials referenced herein.

In an aspect, there is provided a method of predicting disease progression risk in a subject with prostate cancer, the method comprising: a) providing a sample containing RNA and DNA material from tumour cells; b) determining or measuring substantially all of 94 patient features comprising the mRNA, CNA, methylation and clinical features listed for PRONTO-m in Table 6, and some or all reference or control features set forth in Table 6; c) comparing said patient features to the reference or control features; and d) computing a prediction score using a classifier that takes said patient feature values as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

In some embodiments, substantially all of 94 patient biomarkers is all 94 patient biomarkers.

In some embodiments, determining the prediction score comprises classifying the patient tumour into a pathological Gleason Grade Group (GG) class.

In some embodiments, the patient tumour is classified in the pathologic GG2 class if the score is 0.5 or the pathologic GG1 class if the score is <0.5.

In some embodiments, if the patient is classified into the pathologic GG1 class, further comprising managing the patient with active surveillance. In some embodiments, if the patient is classified into the pathologic GG2 class, further comprising treating the patient with surgery, endocrine therapy, chemotherapy, radiotherapy, hormone therapy, gene therapy, thermal therapy, or ultrasound therapy.

The present system and method may be practiced in various embodiments. A suitably configured computer device, and associated communications networks, devices, software and firmware may provide a platform for enabling one or more embodiments as described above. By way of example, FIG. 5 shows a generic computer device 100 that may include a central processing unit (“CPU”) 102 connected to a storage unit 104 and to a random access memory 106. The CPU 102 may process an operating system 101, application program 103, and data 123. The operating system 101, application program 103, and data 123 may be stored in storage unit 104 and loaded into memory 106, as may be required. Computer device 100 may further include a graphics processing unit (GPU) 122 which is operatively connected to CPU 102 and to memory 106 to offload intensive image processing calculations from CPU 102 and run these calculations in parallel with CPU 102. An operator 107 may interact with the computer device 100 using a video display 108 connected by a video interface 105, and various input/output devices such as a keyboard 115, mouse 112, and disk drive or solid state drive 114 connected by an I/O interface 109. In known manner, the mouse 112 may be configured to control movement of a cursor in the video display 108, and to operate various graphical user interface (GUI) controls appearing in the video display 108 with a mouse button. The disk drive or solid state drive 114 may be configured to accept computer readable media 116. The computer device 100 may form part of a network via a network interface 111, allowing the computer device 100 to communicate with other suitably configured data processing systems (not shown). One or more different types of sensors 135 may be used to receive input from various sources.

The present system and method may be practiced on virtually any manner of computer device including a desktop computer, laptop computer, tablet computer or wireless handheld. The present system and method may also be implemented as a computer-readable/useable medium that includes computer program code to enable one or more computer devices to implement each of the various process steps in a method in accordance with the present invention. In case of more than computer devices performing the entire operation, the computer devices are networked to distribute the various steps of the operation. It is understood that the terms computer-readable medium or computer useable medium comprises one or more of any type of physical embodiment of the program code. In particular, the computer-readable/useable medium can comprise program code embodied on one or more portable storage articles of manufacture (e.g. an optical disc, a magnetic disk, a tape, etc.), on one or more data storage portioned of a computing device, such as memory associated with a computer and/or a storage system.

In an aspect, there is provided a computer-implemented method of predicting disease progression risk in a patient with prostate cancer, the method comprising: a) receiving, at at least one processor, data reflecting substantially all of the patient features defined in claim 1 or 7 corresponding to the PRONTO-e or PRONTO-m classifiers regarding a prostate cancer tumor, and some or all reference or control features set forth in Table 6; b) constructing, at at least one processor, a patient profile based on the patient features; c) comparing, at the at least one processor, said patient profile to the reference or control; d) computing, at the at least one processor, a prediction score using a classifier that takes said patient profile as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

In an aspect, there is provided a computer program product for use in conjunction with a general-purpose computer having a processor and a memory connected to the processor, the computer program product comprising a computer readable storage medium having a computer mechanism encoded thereon, wherein the computer program mechanism may be loaded into the memory of the computer and cause the computer to carry out the method of any one of claims 13-15.

In an aspect, there is provided a computer readable medium having stored thereon a data structure for storing the computer program product according to claim 16.

In an aspect, there is provided a device for predicting disease progression risk in a patient with prostate cancer, the device comprising: at least one processor; and electronic memory in communication with the at least one processor, the electronic memory storing processor-executable code that, when executed at the at least one processor, causes the at least one processor to: a) receive data reflecting substantially all of the patient features defined in claim 1 or 7 corresponding to PRONTO-e or PRONTO-m classifiers regarding the prostate cancer tumor, and some or all reference or control features set forth in Table 6; b) compare said patient features to the reference or control features; and c) compute, at the at least one processor, a prediction score using a classifier that takes said patient profile as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

The advantages of the present invention are further illustrated by the following examples. The examples and their particular details set forth herein are presented for illustration only and should not be construed as a limitation on the claims of the present invention.

EXAMPLES

Materials and Methods

Patient Samples:

To train and validate classifiers, radical prostatectomy samples were identified using local electronic medical records at Kingston General Hospital (diagnosis between 1999 and 2012), Montreal General Hospital at McGill University Health Centre (1994-2013) and London Health Sciences Centre (LHSC) (2004-2009). Initial inclusion criteria were (i) reviewed diagnosis of GG1 or GG2 on core biopsy, (ii) underwent radical prostatectomy, and (iii) treatment-naïve prior to surgery. Patients with clinical stage of T3 or higher were excluded. Cases were assigned to either the training cohort or the validation cohort.

For all cases, central pathology review of both diagnostic core biopsies and radical prostatectomies was performed by expert pathologists (FB, MM, DB, TJ). Where possible, DNA and RNA were extracted from punch cores obtained from two areas of the dominant tumor focus (FIG. 1A) [21] enriched for relatively high and low GG regions where present, and using protocols optimized for this approach [22, 23]. All analyses performed were approved by local ethics review panels (Table 3), which allowed a waiver for informed consent. Overall, we collected 633 samples from 333 cases for the training set, and 346 samples from 202 cases for the validation set (see CONSORT data in Table 4).

The clinicopathologic features of the training and validation cohorts are summarized in Table 1. We were 89% powered to validate two classifiers (α=0.01), assuming true positive (TP) rates 0.8 and false positive (FP) rates ≤0.55 [24].

Selection of Candidate Features for Classifiers:

Multiple functional aspects reflecting the biology of GG were interrogated with molecular features at the transcriptomic (mRNA abundance), genomic (DNA copy number alteration, CNA) and epigenomic levels (DNA methylation) (FIG. 1B). A list of 462 molecular features assessing 431 genes/loci (where a gene/locus may be assessed by more than one feature) was assembled following detailed literature review and input from a number of research strands led by members of the study team [25-30] (see Methods; Table 6). We also included four clinical features assessed at diagnosis, and a fifth clinical feature that integrates them into a Cancer of the Prostate Risk Assessment (CAPRA) risk group [31]. In total, we used 467 features for describing tumor samples (Table 6)

Centralized Molecular Profiling:

We employed four molecular diagnostics platforms of which three are currently in clinical use for molecular diagnostics of cancer. The mRNA analysis was performed using the Nanostring N-counter platform [32] with a specific code set developed for this study. CNA analysis was performed both using a multiplex ligation-dependent probe amplification (MLPA)-based assay developed specifically for this project and a custom NanoString copy number codeset [33] [34]. (Ebrahimizadeh et al, submitted manuscript). Finally, epigenetic profiling was performed using methylation-specific polymerase chain reaction (MSP) [26]. All samples in both cohorts were profiled on as many platforms as possible given their RNA and DNA yields.

Development and Validation of Prognostic Classifiers:

Both training and validation data were preprocessed as described in Supplementary Methods. We created a supervised machine learning pipeline (FIG. 1C; Supplementary Methods) to develop a classifier that takes a patient's profile (comprised of feature values) as input and uses pathological prostatectomy GG as the endpoint, where GG1 and GG≥2 cases are the negative and positive gold standards, respectively. Using the training data, >41,000 GG classifiers were evaluated by subjecting selected features to 12 different machine learning algorithms in five-fold cross-validation. Specifically, area under the receiver-operator curve (AUC), TP, FP and true negative (TN) rates were computed for each classifier. This set of metrics was calculated with only the low- or high-grade sample from each case, and we also calculated the mean of the low- and high-grade statistics. We selected two classifiers for validation by prioritizing those with greater TP rates and AUCs from cross-validation.

We validated classifiers by computing statistics as above, and also by randomly selecting one sample (high- or low-grade) per patient in the validation cohort for computing performance statistics, and repeated this process 1000 times. These sampling-based statistics better simulate clinical practice. All statistical analyses were performed using the R software framework (v3.4.3) [35], the machine learning package mlr (v2.15.0) [36] and the plotting package BoutrosLab.plotting.general (v5.9.8) [37].

Ethical Review

All research was performed according to the Tri-Council Policy Statement (TCPS2) and following ethical approval of the study protocol at each participating institute's research ethics board (Table 3).

Selection of Features

CNA Features: MLPA Assay

A multiplexed ligation-dependent probe amplification (MLPA) assay was developed to assess fourteen loci for copy number alterations (CNA; Table 6) previously associated with clinical outcome in prostate cancer (CaP; Ebrahimizadeh et al, submitted manuscript). The loci assayed include the MYC oncogene S[1-3], the PTEN S[4-7], TP53 S[2, 8, 9], CDKN1B S[10, 11], and RB1 S[12, 13] tumor suppressors, loci associated with metastasis such as GABARAPL2 S[13, 14] and PDPK1 S[15, 16], loci associated with maintenance of genomic stability such as RWDD3 S[17-20], GTF2H2 S[21-24] and WRN S[13, 25-27], and genes associated with CaP subtypes: CHD1 S[13, 28, 29], MAP3K7 S[13, 28, 30], NKX3-1 S[13] and PDZD2 S[31, 32].

CNA Features: CPC-GENE NanoString Assay

Using DNA CNA assays, the Canadian Prostate Cancer Genome Network (CPC-GENE) identified an association between percentage of genome alteration and reduced biochemical recurrence-free survival in low- to intermediate-risk CaP patients, and developed a classifier that uses CNA features to predict patient outcome S[33]. A NanoString CNA assay was designed to derive values for those features S[34], and here we used the assay to include 92 CNA features: 85 loci (including 151 genes) and seven additional genes associated with CaP in the literature (Table 6).

mRNA Features:

We generated the mRNA abundance gene panel (for the NanoString RNA assay) by combining gene lists from the following studies:

mRNA Features: CPC-GENE

CPC-GENE performed RNA abundance profiling of samples from intermediate-risk patients S[35] and univariate analysis of these data identified 20 genes associated with poor prognosis. These genes were supplemented with 30 genes identified with similar univariate analysis and predictive modeling of RNA data from Taylor et al S[36].

mRNA Features: Stem Cell Signature

The gene list was derived from “reprogramming” four androgen receptor (AR)+ CaP cell lines (LNCaP, LAPC4, CWR22rv1 and VCaP) to a stem-like phenotype S[37]. Agilent Gene Chip analyses of each cell line revealed transcripts with significant abundance changes between parental and reprogrammed cells. These transcripts were then compared across cell lines to derive a ranked list of 132 commonly changed genes associated with reprogramming. This signature identified propensity for recurrence, metastasis and CaP-specific death as described by S[37]. The top 50 genes on this list were included in the RNA panel.

mRNA Features: Epithelial-to-Mesenchymal Transition (EMT) Signature

Using the GEO2R program and the Benjamini—Hochberg method for multiple testing corrections, gene expression data from PC-3, PC-3M, ALVA-31, RWPE-2-w99 cell lines undergoing invasive growth in 3-dimensional cultures (GEO #GSE19426) S[38] were compared to identify 1669 genes dysregulated in at least three of four cell lines. These genes were cross-referenced to the EMT-associated genes in the SABiosciences qRT-PCR array. The resulting 33 overlapping genes were used as the seed list for network building, using the String v9.1 and GeneMania algorithms S[39, 40]. From the resulting network, 37 key genes, including the common nodal points connecting the pathways, were included in the RNA panel.

mRNA Features: Stromal Influence on Epithelial Growth and Differentation.

A list of 318 genes identified as enriched in embryonic prostate stroma S[41-43] was filtered to enrich for genes also expressed in cancer-associated fibroblasts, and for association with clinical and pathological endpoints (recurrence, CaP death and Gleason score) in four publicly available datasets S[36, 44-46]. A list of 80 genes was created by prioritizing those associated with grade group (GG) and/or recurrence in multiple datasets.

mRNA Features: Tumor Cell Metabolism

Eighty-six candidate genes associated with CaP metabolism were identified through in silico gene network analysis linking signaling pathways of sterol regulatory element binding protein 1 (SREBP1), insulin growth factor (IGF), AR and suppressor of cytokine signaling 1 (SOCS1), using the String v9.1 and GeneMania algorithms S[47]. Expression analysis was performed for these genes by Nanostring nCounter assay on discovery and validation cohorts, each comprised of 32 Gleason pattern 3 and 32 Gleason pattern 4 foci from individual tumors. Univariate analysis using the Mann-Whitney U test (p<0.05) identified 25 differentially expressed genes.

mRNA Features: Prostate Homeostasis

This research strand leveraged benign prostate homeostasis as a model for growth and differentiation by steroid hormones, and dysregulation of these pathways in CaP. Transcripts representing this body of work included FER, PTK2, FLT1, LYN, SRC, JAK1, JAK3, MARK3, STAT3, STAT5A, EDF1, WNT11, ITGAV, ITGA2, and ITGV5.

Methylation and mRNA Features: CpG Island Hypermethylation

Genes (n=14) with CpG island hypermethylation in CaP were identified from the literature and DNA methylation of these genes was assayed using methylation-specific PCR as described S[48] to derive values for these methylation features (Table 6). These genes (except UCHL1) were also added to the RNA panel, along with seven additional epigenetic modifying and regulatory genes: DNMT1, EZH2, HDAC1, HIC1, KCNK2, SRP14 and TERT.

In summary, collating genes from each of these strands resulted in a novel NanoString mRNA panel comprising 342 genes (see Table 6) with additional housekeeping genes (see Supplementary Methods). We used the NanoString assay to measure the abundance of mRNAs from each gene, to derive values for our mRNA features.

Clinical Features

The Cancer of the Prostate Risk Assessment (CAPRA) score is computed with five clinical features: 1) age at diagnosis, 2) PSA at diagnosis in ng/ml, 3) biopsy GG (i.e. clinical GG), 4) clinical T stage and 5) percentage of biopsy cores involved with cancer S[49]. The CAPRA score of a patient can be used in turn to assign a CAPRA risk group (low, intermediate, high), and our candidate prognostic classifiers optionally used this group feature. Alternatively, the first four clinical features can be used directly by the classifiers. If the age at diagnosis was unavailable, we used the age at radical prostatectomy (if available). If PSA at diagnosis was unavailable, we used pre-operative PSA (if available). Biopsy GG1 and GG2 were represented as 0 and 1, respectively, to the classifiers. The clinical T stage was simplified to two possible values, T1 and T2, represented as 0 and 1, respectively, to the classifiers.

Preprocessing Training and Validation Data

mRNA Abundance Data.

To select the normalization method to use, we tested 96 different methods supported by the NanoStringNorm R package (v1.1.22; S[50]), by trying different combinations of parameter values, i.e. Background={none, mean, mean.2sd, max}, CodeCount={none, sum, geo.mean}, SampleContent={none, housekeeping.sum, housekeeping.geo.mean, total.sum, top.mean}, OtherNorm={none, rank.normal}. Otherwise, we used round.values=FALSE, take.log=TRUE and default values for the remaining parameters. To assess each normalization method, we computed several metrics with the resulting normalized data. These metrics include:

• • 1) pass if normalized counts of the low-abundance housekeeping genes are significantly lower than those of the mid-level abundance housekeeping genes and similarly with the mid-abundance genes compared to the high-abundance genes (one-sided Student's t-test P<0.05), fail otherwise
  • 2) the dynamic range measured as the percentage increase in the mean normalized count of the high-abundance housekeeping genes relative to the mean of the low-abundance housekeeping genes
  • 3) the concordance between the normalized counts of control samples replicated across cartridges, where a greater value suggests lesser batch effects
  • 4) the number of non-normal samples, where a sample is non-normal if its distribution of normalized counts across endogenous genes does not pass the Shapiro-Wilk test of normality (FDR-adjusted q<0.1)
  • 5) the number of significant cohort covariates, i.e. genes where the patient origin (Kingston General Hospital/Montreal Hospital at McGill University Health Centre) is a significant covariate in a linear model predicting the normalized count, where GG and biochemical recurrence status are other covariates (FDR-adjusted p<0.1)
  • 6) the correlation between the total normalized count of a sample and the age of its source tissue block
  • 7) the percentage of samples that failed; a sample can fail if:
  • a) normalized count=0 for any housekeeping gene
  • b) after computing Z-scores across housekeeping genes with normalized counts, any |Z|>5
  • c) normalization factor <0.3 or >3, if CodeCount normalization was performed
  • d) the sample has an outlier background level (|Z|>5)
  • e) RNA content value <1, if SampleCount normalization was performed
  • f) the sample has an outlier RNA content value (|Z|>5), if SampleCount normalization was performed
  • g) proportion of missing endogenous genes >0.9, where a gene is missing if the normalized count≤0

Only considering methods that passed metric 1 and had inter-cartridge concordance >0.9 and <10% of training samples failed, we ranked the methods by first ranking by metrics 2-7 separately and then taking the consensus ranking generated with the DECOR method (ConsRank package v2.0.1; S[51]. Based on this ranking, we selected the normalization method with Background=none, CodeCount=none, SampleContent=housekeeping.sum with a target value=5000 (which was roughly estimated based on the training data), and OtherNorm=none.

MLPA CNA Data.

One or two probes targeted each gene and each test sample was assayed in duplicate. For each replicate, the signal from each test probe was divided by the signal from each of the ten reference probes, resulting in a set of seven ratios. A probe was considered positive for a CNA when its 95% confidence interval for the replicate's ratios was outside of the probe's 95% confidence intervals for at least two of the three reference samples (fresh healthy female genome, normal FFPE kidney tissue, normal FFPE breast lymph node tissue) (Promega). The probe was considered positive for a test sample if it was positive for both of its replicates. If there was a discrepancy between the replicates, the probe was considered negative for a CNA. If either of the replicates did not pass quality control (Ebrahimizadeh, submitted manuscript), no CNA status was assigned to the given probe in the given test sample. If all probes for a gene were positive, the gene was considered positive for a CNA in the test sample; if there was a discrepancy, the gene was considered negative; otherwise, no CNA status was assigned. Only deletions were considered for RWDD3, GTF2H2, CHD1, MAP3K7, NKX3-1, WRN, PTEN, CDKN1B, RB1, GABARAPL2 and TP53 genes, while only gains were considered for, MYC, PDPK1 and PDZD2 genes.

NanoStrinq CNA Data

Data was preprocessed as previously described S[34].

Methylation Data

C_q values were computed as described previously S[48]. For a given test sample t and target gene g, we computed the methylation level as follows:

m _t,g,i,j,k,l=(C_q,p,g,i−C_q,p,r,j)−(C_q,t,g,k−C_q,t,r,i)

where

• • p indicates the positive control sample on the same plate as the test sample,
  • r indicates the reference sequence (ALU), and
  • i, j, k, l indicate the replicate numbers.

The normalized methylation level was then defined as:

m _t,g=median_i,j,k,l(m_t,g,i,j,k,l)

A machine learning pipeline for the development of prognostic classifiers

We built a pipeline to exhaustively evaluate different methodologies for the development of a prognostic classifier. Specifically, the pipeline uses supervised machine learning methods to develop a classifier that takes a patient profile as input to predict good or poor prognosis (i.e. testing negative and positive, respectively). In our application, we binarized the GG in prostatectomy specimens (i.e. pathological GG) to define the true class of a patient: patients with only GG1 as negative gold standards and patients with GG2 as positive gold standards (Supplementary FIG. 1).

The pipeline is comprised of four main stages: 1) dataset, 2) partition, 3) feature reduction and 4) cross-validation (FIG. 10).

The first stage focuses on preparing the training dataset. The training dataset includes: a patient-sample by feature matrix (i.e. each row represents a patient profile), and a set of true class values with one value for each sample in the matrix. The pipeline can take input data generated by different platforms. In our application, we have clinical/CAPRA, RNA abundance, MLPA/NanoString CNA and methylation data. For each platform, this stage reduces the dataset to samples that do not have any missing data. If multiple platforms are desired, the dataset is also reduced to samples that have data from each platform of interest. Finally, the invariant features, i.e. features that have the same value across all remaining samples, are removed from the dataset.

The second stage focuses on partitioning the training dataset for repeated cross-validation. The dataset is reduced to only low-grade samples, only high-grade samples, or a randomly selected sample per patient, according to the desired option. By default, this stage prepares for five-fold cross-validation repeated 1000 times, and thus the stage creates 1000 partitionings of the dataset into five equally-sized subsets. For each candidate partitioning, each sample is first randomly assigned to one of the five subsets. If the partitioning is balanced with respect to the true class, biochemical recurrence status (which can be related to the true class in our application), and the origin of the sample since our training samples were obtained from different institutions (i.e. Kingston General Hospital, Montreal Hospital at McGill University Health Centre), the partitioning is retained. Specifically, for each pair of subsets in the partitioning, a two-sided Fisher's exact test is used to test for an association with each trait. If any of the potential associations are significant (p<0.05), another candidate partitioning is generated until a balanced one is obtained.

The third stage focuses on feature reduction. For x-fold cross-validation, each partitioning enables x training subsets. In this stage, invariant features, i.e. features that have the same value across all samples, are removed from each training subset. If desired, each remaining feature will then be tested for a univariate association with the true class (e.g. with a two-sided Mann-Whitney U test). Features with a significant association (e.g. P<0.01 or 0.05) are retained.

The fourth stage performs the repeated x-fold cross-validation with the desired machine learning algorithm using the mlr package v2.15.0 S[52] (FIG. 6). Options for the algorithm (mlr implementation identifier follows in parentheses) are: decision tree (classif.rpart), flexible discriminant analysis (classif.earth), GLM with lasso or elasticnet regularization, cross-validated lambda (classif.cvglmnet), k-nearest neighbour (classif.kknn), linear discriminant analysis (classif.lda), logistic regression (classif.logreg), naive Bayes (classif.naiveBayes), nearest shrunken centroid (classif.pamr), quadratic discriminant analysis (classif.qda), random forest (classif.ranger), regularized discriminant analysis (classif.rda), support vector machine (classif.svm). Regardless of the choice of algorithm, the repeated cross-validation is performed with unweighted samples (i.e. all samples are equally-weighted by default).

For algorithms that support sample weighting, this stage also cross-validates different weightings of the negative/positive gold standard classes: 30%/70%, 40%/60%, 50%/50%, 60%/40%, 70%/30%. Specifically, with a w_n%/(100−w_n)% weighting, each negative and positive sample is assigned a weight of w_n/p_n and (100−w_n)/(1−p_n), respectively, where p_a is the proportion of samples in the negative gold standard class;

thus, the total weight of all negative samples makes up w_n% of the overall total and the total weight of all positive samples makes up (100−w_n)% of the overall total. For all other machine learning algorithm parameters, default values are used.

During cross-validation, a classifier is trained on (x−1) of the x folds with the given machine learning algorithm, dataset (prepared in earlier stages) and sample weighting. If this training fails after three attempts, the pipeline skips to training with the next (x−1) folds of data. If successful, the resulting classifier is tested on the remaining fold of data from two perspectives: i) only the low-grade sample from each case, and ii) only the high-grade sample from each case. For each perspective, the pipeline computes the area under the receiver-operator curve (AUC) averaged across the x folds, and using an operating point of 0.5 (in our application, if a sample's score ≥0.5, the patient is predicted as GG1, otherwise, GG≥2), the true positive (TP), false positive (FP) and true negative (TN) rates with all patients in the x folds. Moreover, for each of these statistics, the pipeline reports the mean of the values from the two perspectives [e.g. AUC_mean=(AUC_low+AUC_high)/2] Finally, the pipeline further summarizes by computing median statistics across the repetitions of cross-validation (e.g. across the 1000 partitionings).

Validation of Grade Group Classifiers PRONTO-e and PRONTO-m

We ran the pipeline to exhaustively test all possible methodologies that it supports, thereby enabling a more thorough search for the optimal methodology. Two main factors went into selecting the methodologies for validation. First, we wanted methodologies that resulted in greater AUC values from cross-validation as they suggest greater overall performance of the corresponding classifiers. Second, we favored greater TP rates (i.e. TP rate ≥0.8) as this prioritized correct classification of the GG≥2 cases, in accordance with our consultations with clinicians who prioritized earlier intervention for these cases at the expense of over-treating some GG1 cases (quantified by the FP rate). The 25 top-performing classifiers have AUCs ranging from 0.772 to 0.790 (FIG. 2) and most of them use either regularized discriminant analysis or support vector machines. PRONTO-m is the only top-25 classifier that satisfies the TP rate constraint (TP rate=0.800, AUC=0.774), and we also selected PRONTO-e (TP rate=0.833, AUC=0.770) for validation. Table 5 describes the methodologies used to generate these two classifiers.

Each selected methodology was then used to train a classifier with the unpartitioned training cohort, restricted to patients with data for the required samples and features. As in cross-validation, we computed the mean AUC, TP and FP rates, where the mean is of the value for only low-grade samples and the value for only high-grade samples. Despite known intra-tumoral heterogeneity S[53], at diagnosis, it is unknown how well the grade of a biopsy sample represents the overall grade of the whole tumor. To better mimic this clinical scenario, for each patient in the validation cohort, one sample was randomly selected, statistics were computing using the representative samples and this process was repeated 1000 times. We computed the median AUC, TP and FP rates across these repetitions (i.e. sampling-based statistics).

Similarity Between Molecular Profiles

In this analysis, we computed the similarity between molecular profiles of samples from the same patient (i.e. the similarity between the low- and high-grade sample profiles), thus, patients with only a single sample were excluded. For all platforms, we only considered profiles that do not have missing values (for any features). For the CNA profiles, the profiles were first restricted to features from the MLPA platform since the validated classifiers only use CNA features from this platform. We defined the pairwise similarity between CNA profiles as the fraction of features where both samples have the same CNA status (i.e. altered or unaltered). For the RNA abundance and methylation profiles, we defined pairwise similarity as the concordance coefficient across the feature values.

Univariate Feature Analysis

For each platform separately, we tested each feature for a univariate association with pathological GG (i.e. GG1 versus GG2). Specifically, we randomly selected one sample per case and then for each feature, used the selected samples to quantify the difference in features values of GG2 versus GG1cases, x(GG≥2)−x(GG1), and estimated the significance of the difference. For the RNA and methylation platforms, we defined x(GG1) and x(GG≥2) as the median feature values for GG1 and GG2 cases, respectively, and the significance was estimated using a two-sided Mann-Whitney test comparing the sets of feature values of GG1 and GG≥2 cases. For the CNA platforms, we defined x(GG1) and x(GG≥2) as the proportions of GG1 and GG≥2 cases, respectively, with an identified CNA, and the significance was estimated using a two-sided proportion test. The p values from the statistical tests were adjusted using the Benjamini-Hochberg method, across all features from the same platform (resulting in q values). The sampling procedure and subsequent computation of statistics were repeated 1000 times, allowing the computation of the median, first and third quartile values across the repetitions. This feature analysis was performed separately with the training and validation data. To estimate the significance of the univariate association of a given feature across both cohorts, we used the weighted-Z method to combine the median q value from each cohort, weighting each q value by the number of cases used to compute it S[54].

Results

Overview of Cohorts/Samples

We successfully generated 954 mRNA, 845 NanoString-CNA, 794 MLPA-CNA, and 847 methylation profiles for samples from 535 prostatectomy cases across the training and validation cohorts. We also generated CAPRA scores for 492 cases.

Development and Validation of GG Classifiers

Classifiers were trained on 333 cases from two sites, reserving 202 cases from a 3^rd site for independent validation (Table 4). Of the >41,000 GG classifiers we evaluated, 718 exhibited AUC≥0.75 with TP and TN rates ≥0.5 (i.e. ≥50% cases in each GG class were correctly predicted). Sensitivity for GG2 was prioritized over specificity because of the clinical need for earlier intervention, resulting in our selection of two top-performing classifiers for validation, PRONTO-e and PRONTO-m (Table 5). For cases with GG>2 samples, both of these classifiers were both trained using only the high-grade sample from that case. Performance statistics for the 25 top-performing classifiers (by AUC) are shown in FIG. 2. PRONTO-e uses 353 features, including 342 mRNA abundance and 11 CNA features (Table 6), and a random forest. PARSE-m uses fewer features (94 in total), but draws from more available categories of data (64 mRNA, 14 CNA, 12 methylation and 4 clinical features; Table 6) and uses a support vector machine. Performance statistics computed with only the low- or high-grade sample from each case, and the mean of the low- and high-grade statistics, are presented in FIGS. 3A-C and Table 2.

Despite reported intra-tumoral heterogeneity in prostate cancer [38] we observed remarkable stability in performance statistics when they were computed with one randomly selected sample per case (FIG. 3D). This process mimics sampling error on biopsy and yielded validation performance characteristics for both classifiers that exceeded those of previously validated biomarkers of adverse pathology [19, 20] (Table 2).

The validated classifiers frequently provided consistent GG classification between paired samples from the same case: 70.8% for PRONTO-e and 73.9% for PRONTO-m indicating a high degree of resistance to sampling error. For PRONTO-e, we observed superior agreement between two samples when both were taken from a GG2 versus GG1 case (FIG. 3E). Moreover, of the concordant cases (n=97), the GG2 subset (n=55) has a significantly greater percentage of correct class predictions (two-sided proportion test p=5.3×10⁻⁴), and the trend was also present for PRONTO-m (FIG. 3F).

Molecular Features of Grade Group

We investigated which molecular features were most strongly associated with GG. By univariate analysis, the abundance of 22 transcripts and methylation at 9 loci showed significant association with GG (adjusted p<0.1, see Methods; FIG. 4). Where cell-type specific expression patterns could be discerned, some transcripts were associated with preferential expression in epithelium or stroma [39]. Similar rates of preferential expression were seen for the stromal and epithelial compartments. Similarly, there were similar rates of positive and negative association of each molecular feature with higher GG. Interestingly, no significant univariate association with GG was identified for CNA features, yet their inclusion in multivariate classifiers of GG improved performance (FIG. 3C).

Multimodal Classifiers Outperform CAPRA in Cross-Validation

The CAPRA score represents the current clinical standard for prostate cancer prognosis and it is computed only with non-molecular features such as age at diagnosis and the GG of the biopsy S[49]. Importantly, both PRONTO-e and PRONTO-m classifiers outperform a CAPRA classifier in cross-validation, with greater TP rates and AUCs (FIG. 3A,C).

GG Classifiers and Intra-Tumoral Heterogeneity

ROC curves computed only with the low-grade or high-grade sample from each case in the validation cohort indicate differences in classifier performance depending on the grade of the sample relative to the grade of whole tumor (FIG. 7A). The ROC curves of the PRONTO-m classifier are more divergent than the curves of the PRONTO-e classifier. The prediction score distributions, for GG1 and GG2 cases separately, are also wider for PRONTO-m versus PRONTO-e (FIG. 7B)

We examined the potential impact of intra-tumoral heterogeneity on the validated classifiers by comparing the input profiles (DNA, RNA) for samples from the same case. We quantified the similarity between the CNA profiles and found that the similarity values are significantly greater for the GG1 versus GG≥2 cases (Mann-Whitney test p=0.023). However, for both CNA and RNA data, the median similarity values are greater than 0.9, regardless of the GG subset (FIG. 8), indicating that these molecular input profiles are quite consistent within cases.

DISCUSSION

Here we report the development of GG classifiers and the validation of the PRONTO-e and PRONTO-m classifiers in an independent patient population. These results suggest that incorporating diverse molecular (e.g. mRNA and CNA) features can add significant value (FIG. 3C). Validation demonstrated that the classifiers effectively discriminate between GG1 and GG2 cases (sampling-based AUCs≥0.786). The high TP (≥0.8) and low false negative (≤0.2) rates (Table 2) suggest significant clinical utility in the early management of CaP. Both PRONTO-e and PRONTO-m represent a marked improvement on current approaches. Three commercially available biomarker tests are designed for use on biopsy tissue to inform management of early CaP at the time of diagnosis [40]. Prolaris uses RNA expression data of cell cycle progression genes in combination with clinical/pathological parameters (Myriad Genetics) and reports risk of ten-year prostate-specific mortality [41]. Given that CaPs are typically diagnosed at the age of 50-65 and the vast majority of deaths occur 20-25 years after diagnosis [42], Prolaris may not be well-suited for decisions around AS. The OncotypeDX prostate (Genomic Health), a 17-gene qPCR-based test, and ProMark (Metamark Genetics), a quantitative in situ proteomic test [22, 43], both use biopsy samples to predict adverse pathology, defined as GG≥3 and/or cancer spread outside of the prostate [19, 20, 22]. Notably, none of the currently available tests accurately classifies GG1 versus GG≥2 cases which leaves these intermediate-risk patients in a grey zone for choosing AS. The addition of the OncotypeDx genomic prostate score (GPS) to the CAPRA clinical and pathologic nomogram very slightly improved the AUC for adverse pathology (AUC=0.67) compared to CAPRA alone (AUC=0.63)[20, 44]. ProMark did somewhat better, with a standalone “favorable pathology” call yielding an AUC of 0.69 at the time of biopsy [19] increasing to 0.75 when used solely on patients classified as favorable risk by NCCN (National Comprehensive Cancer Network) guidelines [2, 45].

Despite limited accuracy in biopsies for detecting GG≥2, the gold standard endpoint for AS, both OncotypeDx and ProMark report resistance to tumor heterogeneity [19, 20]. These results suggest that there are measurable underlying clonal changes that mediate CaP aggressiveness, reflect the GG of the whole tumor and are consistently present across areas of phenotypic tumor heterogeneity [46, 47]. The current work derived and independently validated two novel classifiers of GG that demonstrated resistance to tumor heterogeneity, yielding sampling-based AUCs of 0.799 (PRONTO-e) and 0.786 (PRONTO-m). Both classifiers can detect a GG2 tumor using molecular features of phenotypically low-grade tumor samples (FIG. 3E).

PRONTO-e comprises 353 features divided between mRNA abundance and DNA CNA types. The more compact PARSE-m comprises 94 features divided between mRNA abundance, DNA CNA, and DNA methylation types, and includes pre-surgical clinical and pathologic features (age, clinical stage, and PSA, biopsy GG). Although derived from prostatectomy tissue, for which GG is most accurate, both classifiers are resistant to sampling error and therefore there is a high probability that, when used on biopsy tissue, they will better inform decisions around AS versus clinical management. Work to validate the classifiers with biopsy samples from statistically-powered cohorts is currently underway.

When performed on the same patient, OncotypeDx and Prolaris often yield conflicting recommendations [48]. Nevertheless, the tests have demonstrated the potential to reduce biopsy frequency and overtreatment [40] suggesting more accurate tests have similar, if not better, potential impact. Once the PRONTO-e and PRONTO-m performance is validated in core biopsies, these assays have the potential to dramatically improve this impact. It is relatively simple to model the application of each validated classifier to diagnostic biopsies from 1000 hypothetical men selected for AS, with the assumption that 33% of these men would be upgraded during their AS [49]. A test with performance characteristics similar to PRONTO-m would identify 53.4% of men as positive (for risk of occult GG≥2) and 46.6% as negative. Of those testing positive (534/1000 men), 267 would be TPs and benefit from early repeat biopsy and treatment. Of the 466 men testing negative, only 13.5% (63) would be false negatives. For the 26.7% of all cases with a FP result, we suggest the consequence would be an earlier first AS biopsy, not additional biopsies. The early biopsies for these patients would provide pathological reassurance of low GG disease without additional morbidity. The hypothetical results for PRONTO-e are similar (FIG. 9). Over time, the use of such tests could de-intensify surveillance for a large proportion of patients identified as lower risk and, on a population basis, reduce the numbers of biopsy procedures performed.

The current work establishes PRONTO-e and PRONTO-m as molecular biomarkers of GG that are resistant to sampling error, and therefore likely to perform well in diagnostic biopsies. Further work is needed, and ongoing, to fully validate their clinical performance. Multifocal CaPs represent a potential pitfall for any biopsy test in that that biopsies may sample a secondary low-grade focus while failing to sample the higher grade “dominant” or “index” focus. This phenomenon has been estimated to explain 20-30% of cases upgraded between biopsy and prostatectomy [15, 50]. The performance of the classifiers on biopsy tissue could also be compromised by limiting nucleic acid yields from small biopsy tissue samples. This limitation should be balanced by factors expected to improve performance of the classifiers in biopsies relative to surgical samples, including higher quality nucleic acids observed in biopsy tissue [51] and opportunities to employ more sensitive and precise massively parallel sequencing technologies [52] in the clinical assay.

While several studies have related biopsy classifiers to outcomes after surgery, there is little information linking test results to outcomes for men on AS. Further validation of PRONTO-e and PRONTO-m on biopsies from men on AS is needed. Overall, these results indicate that combining transcriptomic, epigenomic, and genomic features can improve the performance of clinically relevant biomarkers for CaP tissue. This result suggests potential benefits for other biospecimen types (e.g., blood or urine) and tumor sites.

Although preferred embodiments of the invention have been described herein, it will be understood by those skilled in the art that variations may be made thereto without departing from the spirit of the invention or the scope of the appended claims. All documents disclosed herein, including those in the following reference list, are incorporated by reference.

REFERENCE LIST

• 1. Esserman L, Shieh Y, Thompson I. Rethinking screening for breast cancer and prostate cancer. JAMA 2009; 302(15):1685-92.
• 2. Mohler J, Bahnson R R, Boston B, et al. NCCN clinical practice guidelines in oncology: prostate cancer. J Natl Compr Canc Netw 2010; 8(2):162-200.
• 3. Humphrey P A, Moch H, Cubilla A L, et al. The 2016 WHO Classification of Tumours of the Urinary System and Male Genital Organs-Part B: Prostate and Bladder Tumours. Eur Urol 2016; 70(1):106-119.
• 4. Eggener S E, Scardino P T, Walsh P C, et al. Predicting 15-year prostate cancer specific mortality after radical prostatectomy. J Urol 2011; 185(3):869-75.
• 5. Ross H M, Kryvenko O N, Cowan J E, et al. Do adenocarcinomas of the prostate with Gleason score (G S)<1=6 have the potential to metastasize to lymph nodes? Am J Surg Pathol 2012; 36(9):1346-52.
• 6. Kelly R S, Sinnott J A, Rider J R, et al. The role of tumor metabolism as a driver of prostate cancer progression and lethal disease: results from a nested case-control study.

Cancer Metab 2016; 4:22.

• 7. Cutruzzola F, Giardina G, Marani M, et al. Glucose Metabolism in the Progression of Prostate Cancer. Front Physiol 2017; 8:97.
• 8. Levesque C, Nelson P S. Cellular Constituents of the Prostate Stroma: Key Contributors to Prostate Cancer Progression and Therapy Resistance. Cold Spring Harb

Perspect Med 2018; 8(8).

• 9. Beltran H, Hruszkewycz A, Scher H I, et al. The Role of Lineage Plasticity in Prostate Cancer Therapy Resistance. Clin Cancer Res 2019; 25(23):6916-6924.
• 10. Morash C, Tey R, Agbassi C, et al. Active surveillance for the management of localized prostate cancer: Guideline recommendations. Can Urol Assoc J 2015; 9(5-6):171-8.
• 11. Chen R C, Rumble R B, Loblaw D A, et al. Active Surveillance for the Management of Localized Prostate Cancer (Cancer Care Ontario Guideline): American Society of Clinical Oncology Clinical Practice Guideline Endorsement. J Clin Oncol 2016; 34(18):2182-90.
• 12. Sanda M G, Cadeddu J A, Kirkby E, et al. Clinically Localized Prostate Cancer: AUA/ASTRO/SUO Guideline. Part II: Recommended Approaches and Details of Specific Care Options. J Urol 2018; 199(4):990-997.
• 13. Mottet N, van den Bergh E, Briers P, et al. Prostate Cancer Full Text Guidelines, Section 6, “Treatment”. https://uroweb.orq/quideline/prostate-cancer/#6.
• 14. Bullock N, Simpkin A, Fowler S, et al. Pathological upgrading in prostate cancer treated with surgery in the United Kingdom: trends and risk factors from the British Association of Urological Surgeons Radical Prostatectomy Registry. BMC Urol 2019; 19(1):94.
• 15. Epstein J I, Feng Z, Trock B J, et al. Upgrading and downgrading of prostate cancer from biopsy to radical prostatectomy: incidence and predictive factors using the modified

Gleason grading system and factoring in tertiary grades. Eur Urol 2012; 61(5):1019-24.

• 16. Corcoran N M, Hovens C M, Hong M K, et al. Underestimation of Gleason score at prostate biopsy reflects sampling error in lower volume tumours. BJU Int 2012; 109(5):660-4.
• 17. Danneman D, Drevin L, Delahunt B, et al. Accuracy of prostate biopsies for predicting Gleason score in radical prostatectomy specimens: nationwide trends 2000-2012. BJU Int 2017; 119(1):50-56.
• 18. Dijkstra S, Hamid A R, Leyten G H, et al. Personalized management in low-risk prostate cancer: the role of biomarkers. Prostate Cancer 2012; 2012:327104.
• 19. Blume-Jensen P, Berman D M, Rimm D L, et al. Development and clinical validation of an in situ biopsy-based multimarker assay for risk stratification in prostate cancer. Clin Cancer Res 2015; 21(11):2591-600.
• 20. Klein E A, Cooperberg M R, Magi-Galluzzi C, et al. A 17-gene assay to predict prostate cancer aggressiveness in the context of Gleason grade heterogeneity, tumor multifocality, and biopsy undersampling. Eur Urol 2014; 66(3):550-60.
• 21. van der Kwast T H, Amin M B, Billis A, et al. International Society of Urological Pathology (ISUP) Consensus Conference on Handling and Staging of Radical

Prostatectomy Specimens. Working group 2: T2 substaging and prostate cancer volume. Mod Pathol 2011; 24(1):16-25.

• 22. Shipitsin M, Small C, Choudhury S, et al. Identification of proteomic biomarkers predicting prostate cancer aggressiveness and lethality despite biopsy-sampling error. Br J Cancer 2014; 111(6):1201-12.
• 23. Patel P G, Selvarajah S, Boursalie S, et al. Preparation of Formalin-fixed Paraffin-embedded Tissue Cores for both RNA and DNA Extraction. J Vis Exp 2016; 10.3791/54299(114).
• 24. Chow S, Shao J, Wang H. Sample Size Calculations in Clinical Research. London: Chapman and Hall; 2008.
• 25. Lapointe J, Li C, Giacomini C P, et al. Genomic profiling reveals alternative genetic pathways of prostate tumorigenesis. Cancer Res 2007; 67(18):8504-10.
• 26. Patel P G, Wessel T, Kawashima A, et al. A three-gene DNA methylation biomarker accurately classifies early stage prostate cancer. Prostate 2019; 79(14):1705-1714.
• 27. Nouri M, Caradec J, Lubik A A, et al. Therapy-induced developmental reprogramming of prostate cancer cells and acquired therapy resistance. Oncotarget 2017; 8(12):18949-18967.
• 28. Orr B, Riddick A C, Stewart G D, et al. Identification of stromally expressed molecules in the prostate by tag-profiling of cancer-associated fibroblasts, normal fibroblasts and fetal prostate. Oncogene 2012; 31(9):1130-42.
• 29. Lalonde E, Alkallas R, Chua M L K, et al. Translating a Prognostic DNA Genomic Classifier into the Clinic: Retrospective Validation in 563 Localized Prostate Tumors. Eur Urol 2017; 72(1):22-31.
• 30. Rocha J, Zouanat F Z, Zoubeidi A, et al. The Fer tyrosine kinase acts as a downstream interleukin-6 effector of androgen receptor activation in prostate cancer. Mol Cell Endocrinol 2013; 381(1-2):140-9.
• 31. Cooperberg M R, Pasta D J, Elkin E P, et al. The University of California, San Francisco Cancer of the Prostate Risk Assessment score: a straightforward and reliable preoperative predictor of disease recurrence after radical prostatectomy. J Urol 2005; 173(6): 1938-42.
• 32. Bayani J, Yao C Q, Quintayo M A, et al. Molecular stratification of early breast cancer identifies drug targets to drive stratified medicine. NPJ Breast Cancer 2017; 3:3.
• 33. Lacle M M, Kornegoor R, Moelans C B, et al. Analysis of copy number changes on chromosome 16 q in male breast cancer by multiplex ligation-dependent probe amplification. Mod Pathol 2013; 26(11):1461-7.
• 34. Sendorek D H, Lalonde E, Yao C Q, et al. NanoStringNormCNV: pre-processing of NanoString CNV data. Bioinformatics 2018; 34(6):1034-1036.
• 35. Team” R C. R: A Language and Environment for Statistical Computing. In. Vienna, Austria: R Foundation for Statistical Computing; 2017.
• 36. Bischl B, Lang M, Kotthoff L, et al. mlr: Machine Learning in R. Journal of Machine

Learning 2016; 17:1-5.

• 37. P'ng C, Green J, Chong L C, et al. BPG: Seamless, automated and interactive visualization of scientific data. BMC Bioinformatics 2019; 20(1):42.
• 38. Boutros P C, Fraser M, Harding N J, et al. Spatial genomic heterogeneity within localized, multifocal prostate cancer. Nat Genet 2015; 47(7):736-45.
• 39. The Human Protein Atlas. proteinatlas.org (2020 02 07; date last accessed).
• 40. Olleik G, Kassouf W, Aprikian A, et al. Evaluation of New Tests and Interventions for Prostate Cancer Management: A Systematic Review. J Natl Compr Canc Netw 2018; 16(11): 1340-1351.
• 41. Cuzick J, Stone S, Fisher G, et al. Validation of an RNA cell cycle progression score for predicting death from prostate cancer in a conservatively managed needle biopsy cohort. Br J Cancer 2015; 113(3):382-9.
• 42. Albertsen P C, Hanley J A, Fine J. 20-year outcomes following conservative management of clinically localized prostate cancer. JAMA 2005; 293(17):2095-101.
• 43. Saad F, Shipitsin M, Choudhury S, et al. Distinguishing aggressive versus nonaggressive prostate cancer using a novel prognostic proteomics biopsy test, ProMark.

J Clin Oncol 2014; 32(15_suppl):5090-5090.

• 44. Brand T C, Zhang N, Crager M R, et al. Patient-specific Meta-analysis of 2 Clinical Validation Studies to Predict Pathologic Outcomes in Prostate Cancer Using the 17-Gene Genomic Prostate Score. Urology 2016; 89:69-75.
• 45. Mohler J L, Armstrong A J, Bahnson R R, et al. Prostate cancer, Version 3.2012:

featured updates to the NCCN guidelines. J Natl Compr Canc Netw 2012; 10(9):1081-7.

• 46. Boyd L K, Mao X, Lu Y J. The complexity of prostate cancer: genomic alterations and heterogeneity. Nat Rev Urol 2012; 9(11):652-64.
• 47. Sowalsky A G, Ye H, Bubley G J, et al. Clonal progression of prostate cancers from Gleason grade 3 to grade 4. Cancer Res 2013; 73(3):1050-5.
• 48. Alam S, Tortora J, Staff I, et al. Prostate cancer genomics: comparing results from three molecular assays. Can J Urol 2019; 26(3):9758-9762.
• 49. Klotz L. Active surveillance in intermediate-risk prostate cancer. BJU Int 2020; 125(3):346-354.
• 50. Truong M, Slezak J A, Lin C P, et al. Development and multi-institutional validation of an upgrading risk tool for Gleason 6 prostate cancer. Cancer 2013; 119(22):3992-4002.
• 51. Lin D W, Coleman I M, Hawley S, et al. Influence of surgical manipulation on prostate gene expression: implications for molecular correlates of treatment effects and disease prognosis. J Clin Oncol 2006; 24(23):3763-70.
• 52. Dehghani M, Rosenblatt K P, Li L, et al. Validation and Clinical Applications of a Comprehensive Next Generation Sequencing System for Molecular Characterization of Solid Cancer Tissues. Front Mol Biosci 2019; 6:82.
• 53. Clark T G, Bradburn M J, Love S B, et al. Survival analysis part I: basic concepts and first analyses. Br J Cancer 2003; 89(2):232-8.
• 54. Lapointe J, Malhotra S, Higgins J P, et al. hCAP-D3 expression marks a prostate cancer subtype with favorable clinical behavior and androgen signaling signature. Am J Surg Pathol 2008; 32(2):205-9.
• 51. Barros-Silva J D, Ribeiro F R, Rodrigues A, et al. Relative 8 q gain predicts disease-specific survival irrespective of the TMPRSS2-ERG fusion status in diagnostic biopsies of prostate cancer. Genes Chromosomes Cancer 2011; 50(8):662-71.
• S2. Qian J, Hirasawa K, Bostwick D G, et al. Loss of p53 and c-myc overrepresentation in stage T(2-3)N(1-3)M(0) prostate cancer are potential markers for cancer progression. Mod Pathol 2002; 15(1):35-44.
• S3. Zafarana G, Ishkanian A S, Malloff C A, et al. Copy number alterations of c-MYC and PTEN are prognostic factors for relapse after prostate cancer radiotherapy. Cancer 2012; 118(16):4053-62.
• S4. Bramhecha Y M, Rouzbeh S, Guerard K P, et al. The combination of PTEN deletion and 16p13.3 gain in prostate cancer provides additional prognostic information in patients treated with radical prostatectomy. Mod Pathol 2019; 32(1):128-138.
• S5. Choucair K, Ejdelman J, Brimo F, et al. PTEN genomic deletion predicts prostate cancer recurrence and is associated with low A R expression and transcriptional activity. BMC Cancer 2012; 12:543.
• S6. Krohn A, Diedler T, Burkhardt L, et al. Genomic deletion of PTEN is associated with tumor progression and early PSA recurrence in ERG fusion-positive and fusion-negative prostate cancer. Am J Pathol 2012; 181(2):401-12.
• S7. Yoshimoto M, Cunha I W, Coudry R A, et al. FISH analysis of 107 prostate cancers shows that PTEN genomic deletion is associated with poor clinical outcome. Br J Cancer 2007; 97(5):678-85.
• S8. Hamid A A, Gray K P, Shaw G, et al. Compound Genomic Alterations of TP53, PTEN, and RB1 Tumor Suppressors in Localized and Metastatic Prostate Cancer. Eur Urol 2019; 76(1):89-97.
• S9. Kluth M, Harasimowicz S, Burkhardt L, et al. Clinical significance of different types of p53 gene alteration in surgically treated prostate cancer. Int J Cancer 2014; 135(6):1369-80.
• S10. Kluth M, Ahrary R, Hube-Magg C, et al. Genomic deletion of chromosome 12p is an independent prognostic marker in prostate cancer. Oncotarget 2015; 6(29):27966-79.
• S11. Kibel A S, Schutte M, Kern S E, et al. Identification of 12p as a region of frequent deletion in advanced prostate cancer. Cancer Res 1998; 58(24):5652-5.
• S12. Kluth M, Scherzai S, Buschek F, et al. 13 q deletion is linked to an adverse phenotype and poor prognosis in prostate cancer. Genes Chromosomes Cancer 2018; 57(10):504-512.
• S13. Lapointe J, Li C, Giacomini C P, et al. Genomic profiling reveals alternative genetic pathways of prostate tumorigenesis. Cancer Res 2007; 67(18):8504-10.
• S14. Kluth M, Runte F, Barow P, et al. Concurrent deletion of 16q23 and PTEN is an independent prognostic feature in prostate cancer. Int J Cancer 2015; 137(10):2354-63.
• S15. Choucair K A, Guerard K P, Ejdelman J, et al. The 16p13.3 (PDPK1) Genomic Gain in Prostate Cancer: A Potential Role in Disease Progression. Transl Oncol 2012; 5(6):453-60.
• S16. Bramhecha Y M, Guerard K P, Rouzbeh S, et al. Genomic Gain of 16p13.3 in Prostate Cancer Predicts Poor Clinical Outcome after Surgical Intervention. Mol Cancer Res 2018; 16(1):115-123.
• S17. Sun J, Liu W, Adams T S, et al. DNA copy number alterations in prostate cancers: a combined analysis of published CGH studies. Prostate 2007; 67(7):692-700.
• S18. Paris P L, Albertson D G, Alers J C, et al. High-resolution analysis of paraffin-embedded and formalin-fixed prostate tumors using comparative genomic hybridization to genomic microarrays. Am J Pathol 2003; 162(3):763-70.
• S19. Carbia-Nagashima A, Gerez J, Perez-Castro C, et al. RSUME, a small RWD-containing protein, enhances SUMO conjugation and stabilizes HIF-1alpha during hypoxia. Cell 2007; 131(2):309-23.
• S20. Druker J, Liberman A C, Antunica-Noguerol M, et al. RSUME enhances glucocorticoid receptor SUMOylation and transcriptional activity. Mol Cell Biol 2013; 33(11):2116-27.
• S21. Wood R D, Mitchell M, Sgouros J, et al. Human DNA repair genes. Science 2001; 291(5507):1284-9.
• S22. Chymkowitch P, Le May N, Charneau P, et al. The phosphorylation of the androgen receptor by TFIIH directs the ubiquitin/proteasome process. EMBO J 2011; 30(3):468-79.
• S23. Ishkanian A S, Mallof C A, Ho J, et al. High-resolution array CGH identifies novel regions of genomic alteration in intermediate-risk prostate cancer. Prostate 2009; 69(10):1091-100.
• S24. Cheng I, Levin A M, Tai Y C, et al. Copy number alterations in prostate tumors and disease aggressiveness. Genes Chromosomes Cancer 2012; 51(1):66-76.
• S25. Shen J C, Loeb L A. The Werner syndrome gene: the molecular basis of RecQ helicase-deficiency diseases. Trends Genet 2000; 16(5):213-20.
• S26. Mo D, Zhao Y, Balajee A S. Human RecQL4 helicase plays multifaceted roles in the genomic stability of normal and cancer cells. Cancer Lett 2018; 413:1-10.
• S27. Fukasawa S, Kino M, Kobayashi M, et al. Genetic changes in pT2 and pT3 prostate cancer detected by comparative genomic hybridization. Prostate Cancer Prostatic Dis 2008; 11(3):303-10.
• S28. Burkhardt L, Fuchs S, Krohn A, et al. CHD1 is a 5q21 tumor suppressor required for ERG rearrangement in prostate cancer. Cancer Res 2013; 73(9):2795-805.
• S29. Huang S, Gulzar Z G, Salari K, et al. Recurrent deletion of CHD1 in prostate cancer with relevance to cell invasiveness. Oncogene 2012; 31(37):4164-70.
• S30. Rodrigues L U, Rider L, Nieto C, et al. Coordinate loss of MAP3K7 and CHD1 promotes aggressive prostate cancer. Cancer Res 2015; 75(6):1021-34.
• S31. Laitinen V H, Akinrinade O, Rantapero T, et al. Germline copy number variation analysis in Finnish families with hereditary prostate cancer. Prostate 2016; 76(3):316-24.
• S32. Chaib H, Rubin M A, Mucci N R, et al. Activated in prostate cancer: a PDZ domain-containing protein highly expressed in human primary prostate tumors. Cancer Res 2001; 61(6):2390-4.
• S33. Lalonde E, Ishkanian A S, Sykes J, et al. Tumour genomic and microenvironmental heterogeneity for integrated prediction of 5-year biochemical recurrence of prostate cancer: a retrospective cohort study. Lancet Oncol 2014; 15(13):1521-1532.
• S34. Lalonde E, Alkallas R, Chua M L K, et al. Translating a Prognostic DNA Genomic Classifier into the Clinic: Retrospective Validation in 563 Localized Prostate Tumors. Eur Urol 2017; 72(1):22-31.
• S35. Fraser M, Sabelnykova V Y, Yamaguchi T N, et al. Genomic hallmarks of localized, non-indolent prostate cancer. Nature 2017; 541(7637):359-364.
• S36. Taylor B S, Schultz N, Hieronymus H, et al. Integrative genomic profiling of human prostate cancer. Cancer Cell 2010; 18(1):11-22.
• S37. Nouri M, Caradec J, Lubik A A, et al. Therapy-induced developmental reprogramming of prostate cancer cells and acquired therapy resistance. Oncotarget 2017; 8(12):18949-18967.
• S38. Harma V, Virtanen J, Makela R, et al. A comprehensive panel of three-dimensional models for studies of prostate cancer growth, invasion and drug responses. PLoS One 2010; 5(5):e10431.
• S39. Warde-Farley D, Donaldson S L, Comes O, et al. The GeneMANIA prediction server: biological network integration for gene prioritization and predicting gene function. Nucleic Acids Res 2010; 38(Web Server issue):W214-20.
• S40. Franceschini A, Szklarczyk D, Frankild S, et al. STRING v9.1: protein-protein interaction networks, with increased coverage and integration. Nucleic Acids Res 2013; 41(Database issue): D808-15.
• S41. Orr B, Riddick A C, Stewart G D, et al. Identification of stromally expressed molecules in the prostate by tag-profiling of cancer-associated fibroblasts, normal fibroblasts and fetal prostate. Oncogene 2012; 31(9):1130-42.
• S42. Vanpoucke G, Orr B, Grace O C, et al. Transcriptional profiling of inductive mesenchyme to identify molecules involved in prostate development and disease. Genome Biol 2007; 8(10):R213.
• S43. Boufaied N, Nash C, Rochette A, et al. Identification of genes expressed in a mesenchymal subset regulating prostate organogenesis using tissue and single cell transcriptomics. Sci Rep 2017; 7(1):16385.
• S44. Sun Y, Goodison S. Optimizing molecular signatures for predicting prostate cancer recurrence. Prostate 2009; 69(10):1119-27.
• S45. Sboner A, Demichelis F, Calza S, et al. Molecular sampling of prostate cancer: a dilemma for predicting disease progression. BMC Med Genomics 2010; 3:8.
• S46. Mortensen M M, Hoyer S, Lynnerup A S, et al. Expression profiling of prostate cancer tissue delineates genes associated with recurrence after prostatectomy. Sci Rep 2015; 5:16018.
• S47. Roberto D, Selvarajah S, Park P C, et al. Functional validation of metabolic genes that distinguish Gleason 3 from Gleason 4 prostate cancer foci. Prostate 2019; 79(15):1777-1788.
• S48. Patel P G, Wessel T, Kawashima A, et al. A three-gene DNA methylation biomarker accurately classifies early stage prostate cancer. Prostate 2019; 79(14):1705-1714.
• S49. Cooperberg M R, Broering J M, Carroll P R. Risk assessment for prostate cancer metastasis and mortality at the time of diagnosis. J Natl Cancer Inst 2009; 101(12):878-87.
• S50. Waggott D, Chu K, Yin S, et al. NanoStringNorm: an extensible R package for the pre-processing of NanoString mRNA and miRNA data. Bioinformatics 2012; 28(11):1546-8.
• S51. D'Ambrosio A, Mazzeo G, lorio C, et al. A differential evolution algorithm for finding the median ranking under the Kemeny axiomatic approach. Computers & Operations Research 2017; 82:126-138.
• S52. Bischl B, Lang M, Kotthoff L, et al. mlr: Machine Learning in R. Journal of Machine Learning 2016; 17:1-5.
• S53. Boutros P C, Fraser M, Harding N J, et al. Spatial genomic heterogeneity within localized, multifocal prostate cancer. Nat Genet 2015; 47(7):736-45.
• S54. Whitlock M C. Combining probability from independent tests: the weighted Z-method is superior to Fisher's approach. J Evol Biol 2005; 18(5):1368-73.

TABLE 1
Clinicopathologic features of training and validation cohorts.
	Training Cohort	Validation Cohort
	n = 333	n = 202
Age at diagnosis, years:
Median (min, max)	60	(43, 71)	67	(42, 86)
≤65 yrs, n(%)	267	(80.2)	76	(37.6)
>65 yrs, n(%)	65	(19.5)	126	(62.4)
N.D. n(%)	1	(0.3)
Clinical stage, n(%)
T1	56	(16.8)	116	(57.4)
T1c	161	(48.3)
T2	32	(9.6)	76	(37.6)
T2a	41	(12.3)
T2b	9	(2.7)
T2c	4	(1.2)
N.D.	30	(9.0)	12	(5.0)
Pathologic stage, n(%)
pT2	232	(69.7)	168	(83.2)
pT3a	91	(27.3)	31	(15.3)
pT3b	10	(3.0)	3	(1.5)
Biopsy Grade Group, n(%)
GG1 (Gleason 3 + 3 = 6)	204	(61.3)	153	(75.7)
GG2 (Gleason 3 + 4 = 7)	129	(38.7)	49	(24.3)
Prostatectomy Grade
Group, n(%)
1 (Gleason 6)	138	(41.4)	120	(59.4)
2 (Gleason 3 + 4 = 7)	144	(43.2)	74	(36.6)
3 (Gleason 4 + 3 = 7)	45	(13.5)	6	(3.0)
4 and 5 (Gleason 8-10)	6	(0.18)	2	(1.0)
Preoperative PSA
Median (min, max)	6.05	(0.98, 35.56)	5.21	(0.98, 23)
Biochemical recurrence, n(%)
Negative	210	(63.1)	151	(74.8)
Positive	55	(16.5)	19	(9.4)
N.D.	68	(20.4)	32	(15.8)
Margin status, n(%):
Negative	261	(78.4)	177	(87.6)
Positive	72	(21.6)	25	(12.4)
Time to last follow-up, years
Median (95% confidence	6.38	(5.92-7.34)	7.52	(7.18-7.97)
interval)*
*Calculated using the reverse Kaplan-Meier method [53]

TABLE 2
Classifier performance.
Cohort, statistic	Classifier	TPR	FPR	TNR	FNR	AUC
Training,	PRONTO-e	0.833	0.490	0.510	0.167	0.770
low-high meana	PRONTO-m	0.800	0.415	0.585	0.200	0.774
Validation,	PRONTO-e	0.809	0.429	0.571	0.191	0.792
low-high meana	PRONTO-m	0.760	0.262	0.738	0.240	0.818
Validation,	PRONTO-e	0.802	0.403	0.597	0.198	0.799
samp1ing-basedb	PRONTO-m	0.810	0.398	0.602	0.190	0.786
aMean values represent the average computed over the values derived from low-grade and high-grade samples.
bSamp1ing-based statistics provide a better representation of clinical practice (see Methods).
Abbreviations: TPR—true positive rate; FPR—false positive rate; TNR—true negative rate; FNR—false negative rate; AUC—area under the (receiver-operator) curve.

TABLE 3
List of local ethics approvals.
Site                             Approved protocol #
London Health Science Centre Ethics Review 107429
Board
Kingston General Hospital Ethics Review Board 6007088
Montreal University Health Centre Ethics Review 2011-921, 10-115
Board
University of Toronto Ethics Review Board* 31098
*Samples from the collecting hospitals were processed at the Ontario Institute for Cancer Research (OICR) under approval from the University of Toronto ethics board which is the research ethics board of record for OICR.

TABLE 4
CONSORT table for training and validation cohorts.
	Training Cohort	Validation Cohort
Cases with extracted	547	248
samples and/or some
clinical data collected
Cases with central	401	223
pathology review
completeda
Cases meeting	367	207
diagnostic clinical
feature restrictions
(biopsy GG1 or GG2;
cT < T3)b
Cases with ≥1 sample	333	202
with molecular datac
Cases with complete	298	194
clinical data for
CAPRA score
Cases used for	PRONTO-	PRONTO-	PRONTO-	PRONTO-
training/validation of	e	m	e	m
classifiers*	272	235	200	141
aFor 146 cases in the training cohort and 25 cases in the validation cohort, primary core biopsy material was not available for central pathology review leading to exclusion of these cases.
bA further 34 and 16 cases were excluded after pathology review (GG >2, etc).
cFor 34 and 5 cases respectively there was no/incomplete molecular data captured.

TABLE 5
Methodologies for generating classifiers.
	PRONTO-e	PRONTO-m
Data platforms	RNA, CNA	RNA, CNA,
		methylation, clinical
Sample grade (for training)	high-grade only	high-grade only
Partitioning	5-fold, 1000 repeats	5-fold, 1000 repeats
Feature reduction	none	RNA, P < 0.01
		methylation, P < 0.05
Machine learning algorithm	random forest	support vector
		machine
Sample weighting	40%/60%	unweighted

TABLE 6
Candidate features for GG classifiers.
Features sheet
List of candidate features for GG classifiers, corresponding univariate analysis results
(see Supp1ementary Methods), and indicators of membership in the validated GG classifiers.
For binary features, the Training and Validation differences are defined as a difference in proportions.
Column name	Column description
Data type	Data type of feature
Feature	ID of the feature
Symbol	symbol of gene associated with feature
Entrez gene ID	Entrez ID of gene associated with feature
Training difference	difference between feature value of GG ≥ 2 cases and value of GG1 cases, in the training cohort
Validation difference	difference between feature value of GG ≥ 2 cases and value of GG1 cases, in the validation
	cohort
Combination q	q-value for the significance of the univariate association between the feature values and GG; a
	combination of q-values for the training and validation cohorts
PRONTO-e	1 if the feature is used by the PRONTO-e classifier, 0 otherwise
PRONTO-m	1 if the feature is used by the PRONTO-m classifier, 0 otherwise
CNA Feature Comparison sheet
A comparison of CNA features from the MLPA and NanoString assays. For the NanoString assay, 1-3 genes
were collapsed into signature features (thus, a NanoString feature may be associated with multip1e gene rows).
Column name	Column description
MLPA feature	ID of the feature from the MLPA assay, NA if there is no corresponding feature from this assay
NanoString feature	ID of the feature from the NanoString assay, NA if there is no corresponding feature from this
	assay
Symbol	symbol of gene associated with feature
Entrez gene ID	Entrez ID of gene associated with feature
Map location	map location of gene associated with feature
Data			Entrez	Training	Validation	Combination
type	Feature	Symbol	gene ID	difference	difference	q	PRONTO-e	PRONTO-m
clinical	GG	NA	NA	0.3398	0.8871	5.38E−08	0	1
clinical	PSA	NA	NA	1.4700	10.9000	4.91E−07	0	1
clinical	T stage	NA	NA	0.1838	0.1545	4.19E−04	0	1
methylation	UCHL1	UCHL1	7345	0.6083	0.7777	4.25E−03	0	1
clinical	age	NA	NA	1.5000	4.0000	4.50E−03	0	1
RNA	ALDH1A2	ALDH1A2	8854	−0.4713	−0.1725	7.11E−03	1	1
methylation	CCDC181	CCDC181	57821	0.6263	1.5259	2.22E−02	0	1
RNA	ASPN	ASPN	54829	0.2930	0.3270	2.34E−02	1	1
methylation	APC	APC	324	0.2887	0.6910	2.61E−02	0	1
methylation	GSTP1	GSTP1	2950	0.4143	0.6108	2.68E−02	0	1
RNA	BGN	BGN	633	0.3505	0.3690	2.88E−02	1	1
methylation	HOXD3	HOXD3	3232	0.5063	0.4395	2.99E−02	0	1
RNA	SRD5A2	SRD5A2	6716	−0.3710	−0.2230	3.05E−02	1	1
RNA	MT1L	MT1L	4500	−0.3703	−0.5340	3.17E−02	1	1
RNA	FOLH1	FOLH1	2346	0.4300	0.6200	3.51E−02	1	1
methylation	PTGS2	PTGS2	5743	0.4954	1.4763	3.91E−02	0	1
RNA	CACNG4	CACNG4	27092	−0.5223	−0.6250	4.06E−02	1	1
RNA	F5	F5	2153	0.3560	0.6825	4.25E−02	1	1
methylation	ALDH1A2	ALDH1A2	8854	0.2162	0.3359	5.09E−02	0	1
RNA	ITGBL1	ITGBL1	9358	0.3853	0.3780	5.35E−02	1	1
RNA	GLB1L3	GLB1L3	112937	−0.3633	−0.5340	5.38E−02	1	1
RNA	IGF1	IGF1	3479	−0.3095	−0.2105	5.48E−02	1	1
RNA	NCAPD3	NCAPD3	23310	−0.4883	−0.6000	5.69E−02	1	1
RNA	LRRN1	LRRN1	57633	0.3315	0.4240	5.90E−02	1	1
RNA	EPHX2	EPHX2	2053	−0.2625	−0.1920	6.22E−02	1	1
RNA	TOP2A	TOP2A	7153	0.2800	0.1160	6.27E−02	1	1
methylation	ABCB1	ABCB1	5243	0.3470	0.7223	6.80E−02	0	1
RNA	TRAK1	TRAK1	22906	−0.1125	−0.2240	6.95E−02	1	1
RNA	ALDH2	ALDH2	217	−0.2728	−0.1410	7.13E−02	1	1
RNA	BEND3	BEND3	57673	0.1815	0.3360	7.34E−02	1	0
methylation	GSTM2	GSTM2	2946	0.6165	0.6733	9.04E−02	0	1
RNA	MEIS2	MEIS2	4212	−0.2550	−0.1980	9.12E−02	1	1
RNA	EMP2	EMP2	2013	−0.1162	−0.2595	9.32E−02	1	0
RNA	COL3A1	COL3A1	1281	0.2650	0.1900	9.39E−02	1	1
RNA	CPEB3	CPEB3	22849	−0.1388	−0.1430	9.99E−02	1	1
RNA	GNE	GNE	10020	−0.2030	−0.1270	1.06E−01	1	1
methylation	GAS6	GAS6	2621	0.2714	0.0757	1.07E−01	0	1
RNA	TACC2	TACC2	10579	−0.1335	−0.1090	1.18E−01	1	0
RNA	ATF3	ATF3	467	−0.6800	−0.3600	1.21E−01	1	0
RNA	GSK3B	GSK3B	2932	0.0833	0.0710	1.21E−01	1	0
RNA	CDK1	CDK1	983	0.2365	0.1050	1.24E−01	1	1
RNA	KHDRBS3	KHDRBS3	10656	0.1973	0.2410	1.27E−01	1	1
RNA	NUCB1	NUCB1	4924	−0.0610	−0.1300	1.27E−01	1	0
RNA	GSTM2	GSTM2	2946	−0.2595	−0.2500	1.36E−01	1	1
RNA	MYLK	MYLK	4638	−0.1600	−0.1200	1.37E−01	1	1
RNA	ZNF334	ZNF334	55713	−0.1820	−0.2670	1.40E−01	1	1
RNA	ANXA1	ANXA1	301	−0.1900	−0.2200	1.40E−01	1	1
RNA	LENG9	LENG9	94059	−0.3418	−0.0530	1.41E−01	1	1
RNA	CNN1	CNN1	1264	−0.2680	−0.1760	1.44E−01	1	1
methylation	RASSF1	RASSF1	11186	0.3251	0.2677	1.45E−01	0	0
RNA	GNPTAB	GNPTAB	79158	0.1545	0.2490	1.50E−01	1	1
RNA	RBM24	RBM24	221662	−0.3025	−0.1640	1.52E−01	1	0
RNA	GDF15	GDF15	9518	−0.4450	−0.2700	1.54E−01	1	1
RNA	NXF1	NXF1	10482	−0.0920	−0.0650	1.60E−01	1	1
RNA	ACAD8	ACAD8	27034	−0.1982	−0.2560	1.68E−01	1	0
RNA	SEMA4G	SEMA4G	57715	−0.2465	−0.6010	1.68E−01	1	0
RNA	MYBPC1	MYBPC1	4604	−0.2540	−0.5950	1.81E−01	1	0
RNA	TGFB3	TGFB3	7043	−0.2035	−0.2900	1.81E−01	1	1
RNA	KRT14	KRT14	3861	−0.4973	−0.3500	1.85E−01	1	1
RNA	GMNN	GMNN	51053	0.2433	0.1630	1.86E−01	1	0
RNA	MMP9	MMP9	4318	−0.3330	−0.0850	1.88E−01	1	0
RNA	AZGP1	AZGP1	563	−0.2100	−0.3100	1.90E−01	1	0
RNA	RPTOR	RPTOR	57521	−0.0885	−0.0690	1.90E−01	1	0
RNA	HELB	HELB	92797	−0.1753	−0.1170	1.98E−01	1	0
RNA	CTHRC1	CTHRC1	115908	0.2730	0.3100	1.99E−01	1	1
RNA	ITGA2	ITGA2	3673	−0.1795	−0.2140	1.99E−01	1	1
RNA	TWIST1	TWIST1	7291	0.2388	0.3380	2.00E−01	1	0
RNA	PARM1	PARM1	25849	−0.1745	−0.1650	2.02E−01	1	1
RNA	PI15	PI15	51050	−0.4115	−0.2680	2.03E−01	1	1
RNA	SNAI2	SNAI2	6591	−0.1530	−0.1890	2.04E−01	1	1
RNA	GSTP1	GSTP1	2950	−0.1908	−0.1350	2.05E−01	1	1
RNA	NOTCH3	NOTCH3	4854	0.1227	0.1660	2.08E−01	1	1
RNA	DIS3L2	DIS3L2	129563	−0.0500	−0.0550	2.14E−01	1	1
RNA	ANG	ANG	283	−0.1720	−0.1780	2.16E−01	1	0
RNA	NTRK3	NTRK3	4916	−0.1090	−0.4810	2.17E−01	1	0
RNA	JAK2	JAK2	3717	−0.0635	−0.1090	2.20E−01	1	0
RNA	CFC1	CFC1	55997	−0.2718	−0.4760	2.23E−01	1	0
RNA	CAPN6	CAPN6	827	−0.2050	−0.3080	2.25E−01	1	1
RNA	SMAD4	SMAD4	4089	−0.0320	−0.0690	2.29E−01	1	1
RNA	THBS2	THBS2	7058	0.1873	0.2460	2.29E−01	1	1
RNA	CENPF	CENPF	1063	0.2213	0.2430	2.30E−01	1	1
RNA	TXNL4B	TXNL4B	54957	−0.0940	−0.0515	2.35E−01	1	0
RNA	COL1A1	COL1A1	1277	0.1935	0.1630	2.39E−01	1	1
RNA	STAT5A	STAT5A	6776	−0.0675	−0.0980	2.44E−01	1	0
RNA	THY1	THY1	7070	0.2563	0.2190	2.46E−01	1	1
RNA	NDRG1	NDRG1	10397	0.1950	0.1650	2.52E−01	1	1
RNA	NRP1	NRP1	8829	0.1135	0.2385	2.58E−01	1	0
RNA	VGLL3	VGLL3	389136	−0.2385	−1.0490	2.60E−01	1	0
RNA	AOX1	AOX1	316	−0.1730	−0.2200	2.62E−01	1	0
RNA	PNRC1	PNRC1	10957	−0.0700	−0.0400	2.64E−01	1	0
RNA	P3H2	P3H2	55214	−0.2465	−0.5010	2.66E−01	1	1
methylation	HAPLN3	HAPLN3	145864	0.2527	0.2856	2.69E−01	0	1
RNA	NOX4	NOX4	50507	0.2003	0.1290	2.75E−01	1	1
RNA	FZD7	FZD7	8324	−0.1020	−0.1710	2.83E−01	1	0
RNA	SWT1	SWT1	54823	−0.0685	−0.1620	2.88E−01	1	0
methylation	SEPT9	SEPT9	10801	0.3572	0.9878	2.94E−01	0	0
RNA	NETO2	NETO2	81831	0.2105	0.3280	2.99E−01	1	0
RNA	CX3CL1	CX3CL1	6376	−0.2910	−0.0430	3.00E−01	1	1
CN (MLPA)	MAP3K7	MAP3K7	6885	0.0782	0.1157	3.01E−01	0	1
RNA	DNAH8	DNAH8	1769	0.1765	0.3460	3.04E−01	1	0
RNA	COL6A2	COL6A2	1292	−0.1000	−0.1300	3.09E−01	1	0
RNA	PTPRM	PTPRM	5797	−0.1388	−0.1580	3.16E−01	1	0
RNA	SLC15A2	SLC15A2	6565	−0.1275	−0.4885	3.20E−01	1	0
RNA	INHBA	INHBA	3624	0.1795	0.1475	3.21E−01	1	1
RNA	PSTK	PSTK	118672	−0.0885	−0.0880	3.21E−01	1	0
RNA	KRT15	KRT15	3866	−0.3838	−0.6370	3.25E−01	1	1
RNA	WHSC1	WHSC1	7468	0.0375	0.0820	3.26E−01	1	0
RNA	GSC	GSC	145258	0.0953	0.0905	3.28E−01	1	0
CN (MLPA)	WRN	WRN	7486	0.1070	0.0978	3.28E−01	1	1
RNA	NUDT15	NUDT15	55270	−0.1040	−0.0680	3.31E−01	1	1
RNA	RBPMS	RBPMS	11030	−0.1043	−0.1010	3.33E−01	1	0
RNA	ZSCAN29	ZSCAN29	146050	−0.0840	−0.0740	3.39E−01	1	0
RNA	KRT10	KRT10	3858	−0.1035	−0.3535	3.39E−01	1	0
RNA	SMC4	SMC4	10051	0.1118	0.0860	3.50E−01	1	0
RNA	SHB	SHB	6461	−0.0968	−0.0840	3.62E−01	1	0
RNA	PDLIM7	PDLIM7	9260	−0.0805	−0.0580	3.65E−01	1	0
RNA	ADAM33	ADAM33	80332	−0.1620	−0.2410	3.67E−01	1	0
RNA	SLC45A3	SLC45A3	85414	−0.1025	−0.2000	3.71E−01	1	1
RNA	MCM4	MCM4	4173	0.0640	0.0195	3.75E−01	1	0
RNA	CDKN1B	CDKN1B	1027	−0.0435	−0.1005	3.77E−01	1	0
RNA	KRT5	KRT5	3852	−0.4478	−0.7290	3.77E−01	1	1
RNA	HIC1	HIC1	3090	−0.2030	−0.0140	3.79E−01	1	0
RNA	FHL1	FHL1	2273	−0.0960	−0.3490	3.82E−01	1	0
RNA	SLC7A8	SLC7A8	23428	−0.0910	−0.2870	3.83E−01	1	0
RNA	IL6ST	IL6ST	3572	−0.1150	−0.0700	3.84E−01	1	0
RNA	MYO6	MYO6	4646	0.0550	0.4355	3.90E−01	1	0
methylation	AOX1	AOX1	316	0.3190	0.1399	3.98E−01	0	1
RNA	ZEB1-AS1	ZEB1-AS1	220930	0.0990	0.1070	3.98E−01	1	0
RNA	C1QTNF5	C1QTNF5	114902	0.1365	0.1120	4.08E−01	1	0
RNA	MAP3K7	MAP3K7	6885	−0.0565	−0.0690	4.08E−01	1	0
RNA	AUTS2	AUTS2	26053	−0.0600	−0.1520	4.11E−01	1	0
RNA	ERBB2	ERBB2	2064	−0.0280	−0.1290	4.11E−01	1	0
RNA	PTGS2	PTGS2	5743	−0.2333	0.0055	4.12E−01	1	0
RNA	SATB1	SATB1	6304	−0.1098	−0.0460	4.13E−01	1	0
RNA	RET	RET	5979	−0.1478	−0.5960	4.17E−01	1	0
RNA	AHNAK	AHNAK	79026	−0.0350	−0.1300	4.20E−01	1	0
RNA	NPR3	NPR3	4883	−0.1688	−0.4500	4.20E−01	1	0
RNA	SMPDL3A	SMPDL3A	10924	−0.0243	−0.2220	4.29E−01	1	0
RNA	PTK2	PTK2	5747	0.0625	0.0480	4.32E−01	1	0
RNA	EZH2	EZH2	2146	0.0705	0.0460	4.33E−01	1	1
RNA	CASP8AP2	CASP8AP2	9994	−0.0515	−0.0480	4.40E−01	1	0
RNA	CEBPD	CEBPD	1052	−0.3000	−0.0800	4.41E−01	1	0
RNA	COLGALT2	COLGALT2	23127	−0.0910	−0.1440	4.45E−01	1	0
RNA	SLC1A5	SLC1A5	6510	−0.0165	−0.1860	4.45E−01	1	0
RNA	RABGAP1L	RABGAP1L	9910	−0.0495	−0.0650	4.47E−01	1	0
RNA	PDPK1	PDPK1	5170	0.0365	0.0540	4.50E−01	1	0
RNA	BNC2	BNC2	54796	−0.0960	−0.1140	4.55E−01	1	0
RNA	EGF	EGF	1950	−0.1437	−0.3050	4.55E−01	1	0
RNA	SPARC	SPARC	6678	0.1500	0.0700	4.62E−01	1	1
RNA	TP63	TP63	8626	−0.1933	−0.4570	4.68E−01	1	1
RNA	CAV2	CAV2	858	−0.1268	−0.2410	4.71E−01	1	0
RNA	CCNE2	CCNE2	9134	0.1520	0.0280	4.77E−01	1	1
RNA	MYB	MYB	4602	−0.1290	−0.1130	4.78E−01	1	0
RNA	SOX2	SOX2	6657	−0.2158	−0.2680	4.78E−01	1	0
RNA	WNT5A	WNT5A	7474	0.3345	0.0335	4.80E−01	1	0
RNA	MEG3	MEG3	55384	−0.0060	−0.1940	4.82E−01	1	0
RNA	CYP19A1	CYP19A1	1588	−0.1233	−0.0555	4.93E−01	1	0
RNA	LYN	LYN	4067	0.0633	0.1020	4.94E−01	1	0
RNA	PHF1	PHF1	5252	−0.0495	−0.1200	4.97E−01	1	0
RNA	CDKN1A	CDKN1A	1026	−0.2055	−0.1105	5.01E−01	1	0
RNA	PKP3	PKP3	11187	0.0660	0.1100	5.03E−01	1	0
RNA	FMO5	FMO5	2330	−0.1490	−0.1520	5.04E−01	1	0
RNA	FAM111A	FAM111A	63901	−0.1300	−0.1730	5.11E−01	1	0
RNA	CCNA2	CCNA2	890	0.1045	0.0405	5.15E−01	1	0
RNA	KDR	KDR	3791	−0.0928	−0.1820	5.20E−01	1	0
RNA	SFRP2	SFRP2	6423	0.1678	−0.0240	5.21E−01	1	1
RNA	TNFAIP2	TNFAIP2	7127	−0.1095	−0.0340	5.21E−01	1	0
RNA	OVGP1	OVGP1	5016	−0.0907	−0.0865	5.24E−01	1	0
RNA	KRT8	KRT8	3856	−0.0750	−0.0900	5.25E−01	1	0
RNA	TRIM29	TRIM29	23650	−0.0848	−0.4220	5.33E−01	1	0
RNA	LOX	LOX	4015	0.0925	0.0760	5.34E−01	1	0
RNA	DGCR8	DGCR8	54487	−0.0720	−0.0040	5.37E−01	1	0
RNA	METTL7A	METTL7A	25840	−0.0945	−0.1210	5.48E−01	1	0
RNA	RB1	RB1	5925	−0.0530	−0.0540	5.52E−01	1	0
RNA	SRP14	SRP14	6727	0.0940	0.0190	5.53E−01	1	0
RNA	PCSK6.conser	PCSK6	5046	0.0392	0.1700	5.56E−01	1	0
RNA	CPNE4	CPNE4	131034	0.1330	0.1610	5.57E−01	1	0
RNA	SLC5A12	SLC5A12	159963	−0.1050	−0.1760	5.60E−01	1	0
RNA	H2AFX	H2AFX	3014	0.0473	0.0730	5.65E−01	1	0
RNA	MPDZ	MPDZ	8777	−0.0370	−0.0040	5.66E−01	1	0
RNA	AMACR	AMACR	23600	0.1750	0.3100	5.67E−01	1	0
RNA	RMI2	RMI2	116028	0.1350	0.0770	5.67E−01	1	0
RNA	CCDC181	CCDC181	57821	−0.1353	−0.2060	5.73E−01	1	0
RNA	UNC119	UNC119	9094	0.0425	0.0420	5.77E−01	1	0
RNA	CAMK2N1	CAMK2N1	55450	0.1360	0.0690	5.77E−01	1	0
RNA	HAPLN3	HAPLN3	145864	−0.1400	0.0670	5.78E−01	1	0
RNA	CCNB2	CCNB2	9133	0.1740	0.1060	5.78E−01	1	0
RNA	KCNS3	KCNS3	3790	0.1270	0.0250	5.80E−01	1	0
RNA	EI24	EI24	9538	0.0490	0.0520	5.81E−01	1	0
RNA	CDH1	CDH1	999	−0.0700	−0.1100	5.83E−01	1	0
RNA	FER	FER	2241	−0.0360	−0.0460	5.88E−01	1	0
RNA	MYC	MYC	4609	−0.0530	0.1740	5.91E−01	1	0
RNA	GDAP1	GDAP1	54332	0.0480	0.1715	5.95E−01	1	0
RNA	IL6	IL6	3569	−0.2458	−0.0040	6.04E−01	1	0
RNA	DRD4	DRD4	1815	−0.0900	−0.0920	6.07E−01	1	0
RNA	ABCB1	ABCB1	5243	−0.1068	0.0110	6.09E−01	1	0
RNA	SMAD3	SMAD3	4088	−0.0455	−0.0320	6.09E−01	1	0
RNA	PGRMC1	PGRMC1	10857	−0.0660	−0.0555	6.13E−01	1	0
RNA	JAK1	JAK1	3716	−0.0325	−0.0180	6.16E−01	1	0
RNA	IGFBP3	IGFBP3	3486	0.2030	0.1960	6.16E−01	1	0
RNA	SOX9	SOX9	6662	−0.2365	0.0610	6.16E−01	1	0
RNA	IGF2	IGF2	3481	−0.0840	−0.2315	6.20E−01	1	0
RNA	COL1A2	COL1A2	1278	0.0842	0.0770	6.21E−01	1	1
RNA	FAM114A1	FAM114A1	92689	0.0330	−0.0440	6.22E−01	1	0
RNA	JAG1	JAG1	182	0.0835	0.0360	6.25E−01	1	0
RNA	CISH	CISH	1154	−0.1163	0.0320	6.25E−01	1	0
RNA	RASSF1	RASSF1	11186	−0.0820	−0.0060	6.29E−01	1	0
RNA	APC	APC	324	−0.0475	0.0150	6.30E−01	1	0
RNA	NKX3-1	NKX3-1	4824	0.0650	0.0900	6.37E−01	1	0
RNA	TMPRSS2	TMPRSS2	7113	−0.1100	−0.2300	6.39E−01	1	0
RNA	KRT18	KRT18	3875	−0.0600	−0.0900	6.43E−01	1	0
RNA	TSHR	TSHR	7253	−0.1220	−0.2970	6.46E−01	1	0
CN (MLPA)	PTEN	PTEN	5728	0.0395	0.0886	6.51E−01	0	1
RNA	SLC6A14	SLC6A14	11254	−0.0888	−0.1850	6.52E−01	1	0
RNA	MRC2	MRC2	9902	−0.0778	0.0110	6.53E−01	1	0
RNA	SQRDL	SQRDL	58472	−0.0770	−0.0210	6.56E−01	1	0
RNA	NCOA4	NCOA4	8031	−0.1200	−0.0100	6.57E−01	1	0
RNA	NEFH	NEFH	4744	−0.1215	−0.2280	6.58E−01	1	0
RNA	ADAMTS18	ADAMTS18	170692	−0.0225	−0.2400	6.63E−01	1	0
RNA	PIK3R3	PIK3R3	8503	0.0623	0.0140	6.66E−01	1	0
RNA	SERPINB5	SERPINB5	5268	−0.1920	−0.1640	6.67E−01	1	0
RNA	COL5A2	COL5A2	1290	0.0695	0.0280	6.69E−01	1	1
RNA	COL9A2	COL9A2	1298	0.3450	0.3190	6.69E−01	1	0
RNA	CEP250	CEP250	11190	−0.0695	0.0755	6.78E−01	1	0
RNA	CCNT2-AS1	CCNT2-AS1	100129961	0.0657	−0.0140	6.78E−01	1	0
CN (MLPA)	NKX3-1	NKX3-1	4824	0.0570	0.0972	6.79E−01	1	1
RNA	CCDC8	CCDC8	83987	−0.0798	−0.1030	6.80E−01	1	0
RNA	GSN	GSN	2934	−0.0150	−0.0400	6.80E−01	1	0
RNA	DFNA5	DFNA5	1687	−0.0585	−0.1300	6.82E−01	1	0
RNA	PRIM2	PRIM2	5558	−0.0973	−0.1460	6.82E−01	1	0
RNA	HDAC1	HDAC1	3065	0.0450	0.0900	6.84E−01	1	0
RNA	CAT	CAT	847	−0.0210	−0.0980	6.86E−01	1	0
RNA	RAP1GAP	RAP1GAP	5909	0.0195	0.1225	6.97E−01	1	0
RNA	FRZB	FRZB	2487	0.2205	−0.0760	7.01E−01	1	0
RNA	ROBO1	ROBO1	6091	−0.0833	−0.0135	7.02E−01	1	0
RNA	IL27RA	IL27RA	9466	−0.1253	0.0510	7.03E−01	1	0
RNA	GAS6	GAS6	2621	0.0153	−0.0940	7.03E−01	1	0
RNA	TFDP1	TFDP1	7027	0.0380	0.0440	7.10E−01	1	0
RNA	HSD17B4	HSD17B4	3295	0.0393	0.0205	7.22E−01	1	0
RNA	GAS2L3	GAS2L3	283431	0.1388	−0.0250	7.24E−01	1	0
RNA	NRN1	NRN1	51299	0.1480	−0.0500	7.26E−01	1	0
RNA	COL6A3	COL6A3	1293	−0.0205	−0.0835	7.27E−01	1	0
RNA	CIS	CIS	716	0.0043	−0.1400	7.29E−01	1	0
RNA	HKR1	HKR1	284459	−0.0613	0.0235	7.31E−01	1	0
RNA	SPINK1	SPINK1	6690	−0.3183	−0.5900	7.40E−01	1	0
RNA	PCSK6	PCSK6	5046	0.0107	0.1280	7.45E−01	1	0
CN (MLPA)	GABARAPL2	GABARAPL2	11345	0.0683	0.0354	7.45E−01	1	1
RNA	ITGB5	ITGB5	3693	−0.0330	−0.0630	7.48E−01	1	0
RNA	RPGR	RPGR	6103	−0.0185	−0.0410	7.53E−01	1	0
RNA	PTEN.UTR	PTEN	5728	−0.0323	−0.0390	7.58E−01	1	0
RNA	BHLHE22	BHLHE22	27319	−0.0712	0.0280	7.60E−01	1	0
RNA	VIM	VIM	7431	−0.0200	−0.0300	7.64E−01	1	0
RNA	PLEKHH2	PLEKHH2	130271	−0.0785	−0.0920	7.66E−01	1	0
RNA	UBE2L6	UBE2L6	9246	0.1018	0.0090	7.68E−01	1	0
RNA	CTSF	CTSF	8722	−0.0235	−0.0530	7.72E−01	1	0
CN (MLPA)	MYC	MYC	4609	0.0504	0.0241	7.73E−01	1	1
RNA	HSPA1B	HSPA1B	3304	0.0550	−0.0200	7.82E−01	1	0
RNA	RAMP1	RAMP1	10267	0.0683	−0.1090	7.85E−01	1	0
RNA	CBLB	CBLB	868	−0.0285	−0.0120	7.88E−01	1	0
RNA	M1PH	M1PH	79083	0.0485	−0.0245	7.96E−01	1	0
RNA	TYRO3	TYRO3	7301	−0.0485	−0.0320	7.97E−01	1	0
RNA	HEXDC	HEXDC	284004	0.0355	−0.0140	8.02E−01	1	0
RNA	ITGAV	ITGAV	3685	−0.0185	−0.0540	8.03E−01	1	0
RNA	PTEN	PTEN	5728	−0.0275	−0.0225	8.08E−01	1	0
RNA	CYP3A4	CYP3A4	1576	0.0155	−0.0620	8.08E−01	1	0
RNA	HIF1A	HIF1A	3091	0.0200	0.1100	8.09E−01	1	0
RNA	DHCR7	DHCR7	1717	0.0305	−0.1025	8.09E−01	1	0
RNA	ENPEP	ENPEP	2028	−0.0570	0.0480	8.09E−01	1	0
RNA	TP53	TP53	7157	−0.0410	0.0050	8.09E−01	1	0
RNA	SULT2A1	SULT2A1	6822	−0.0322	−0.0240	8.11E−01	1	0
RNA	RXFP1	RXFP1	59350	0.0245	−0.1260	8.13E−01	1	0
RNA	GPI	GPI	2821	−0.0100	−0.0040	8.18E−01	1	0
RNA	DNMT1	DNMT1	1786	−0.0025	−0.0940	8.18E−01	1	0
RNA	EFNA4	EFNA4	1945	0.0030	0.0590	8.18E−01	1	0
RNA	ANGPT2	ANGPT2	285	−0.0023	−0.1375	8.18E−01	1	0
RNA	PTPN1	PTPN1	5770	−0.0530	−0.0380	8.19E−01	1	0
RNA	AMD1	AMD1	262	−0.0350	0.1200	8.19E−01	1	0
RNA	HOXD3	HOXD3	3232	0.0040	0.1880	8.20E−01	1	0
RNA	RAB27A	RAB27A	5873	−0.0615	−0.1270	8.20E−01	1	0
RNA	ADGRG6	ADGRG6	57211	0.0430	−0.1495	8.20E−01	1	0
RNA	ERICH5	ERICH5	203111	0.0907	0.0340	8.20E−01	1	0
RNA	PDE1A	PDE1A	5136	0.0185	0.1065	8.22E−01	1	0
RNA	NOTCH1	NOTCH1	4851	−0.0345	0.0190	8.23E−01	1	0
RNA	STC1	STC1	6781	0.0200	0.0810	8.24E−01	1	0
RNA	RASSF8	RASSF8	11228	0.0135	0.0570	8.26E−01	1	0
RNA	ABCA3	ABCA3	21	0.0370	−0.0200	8.28E−01	1	0
RNA	TGFA	TGFA	7039	−0.0130	0.0170	8.30E−01	1	0
RNA	HAND2	HAND2	9464	−0.0713	−0.0640	8.32E−01	1	0
RNA	INPP4A	INPP4A	3631	−0.0065	0.0300	8.33E−01	1	0
RNA	ZEB2	ZEB2	9839	−0.0670	0.0490	8.35E−01	1	0
RNA	C18orf21	C18orf21	83608	0.0257	0.0160	8.36E−01	1	0
RNA	NME5	NME5	8382	−0.0585	0.0185	8.38E−01	1	0
RNA	STARD10	STARD10	10809	−0.0728	−0.1680	8.38E−01	1	0
RNA	B4GAT1	B4GAT1	11041	−0.0067	−0.0320	8.40E−01	1	0
RNA	MDM2	MDM2	4193	−0.0145	0.0670	8.42E−01	1	0
RNA	KLK3	KLK3	354	−0.0600	−0.1000	8.43E−01	1	0
RNA	KRT1	KRT1	3848	0.0458	−0.1290	8.44E−01	1	0
RNA	THSD7A	THSD7A	221981	−0.0330	−0.0045	8.44E−01	1	0
RNA	NDC1	NDC1	55706	−0.0050	0.0090	8.45E−01	1	0
RNA	ROBO2	ROBO2	6092	0.1035	−0.0180	8.47E−01	1	0
RNA	GABARAPL2	GABARAPL2	11345	0.0150	−0.0500	8.47E−01	1	0
RNA	FLT1	FLT1	2321	0.0835	−0.0295	8.49E−01	1	0
RNA	EBF3	EBF3	253738	0.0340	0.0605	8.50E−01	1	0
RNA	MUC1	MUC1	4582	−0.0685	−0.0850	8.52E−01	1	0
RNA	WWOX	WWOX	51741	−0.0505	−0.0020	8.53E−01	1	0
RNA	AP1S2	AP1S2	8905	0.0977	−0.0130	8.56E−01	1	0
RNA	SRC	SRC	6714	−0.0388	−0.0330	8.57E−01	1	0
RNA	NOTCH4	NOTCH4	4855	−0.0205	−0.0330	8.58E−01	1	0
RNA	CALD1	CALD1	800	−0.0300	0.0100	8.58E−01	1	0
CN (NanoString)	sig2	NA	NA	0.0850	0.0927	8.59E−01	0	0
RNA	AR	AR	367	−0.0540	−0.0310	8.59E−01	1	0
RNA	HMOX1	HMOX1	3162	0.0102	0.0180	8.62E−01	1	0
RNA	TGFBR2	TGFBR2	7048	0.0158	0.0020	8.63E−01	1	0
RNA	PHACTR2	PHACTR2	9749	−0.0060	0.0075	8.63E−01	1	0
RNA	CDK4	CDK4	1019	0.0213	0.0310	8.63E−01	1	0
RNA	SEPT9	SEPT9	10801	0.0165	0.0860	8.65E−01	1	0
RNA	ZEB1	ZEB1	6935	0.0270	0.0700	8.67E−01	1	0
RNA	GFRA3	GFRA3	2676	−0.0583	−0.1195	8.68E−01	1	0
RNA	SMS	SMS	6611	−0.0975	−0.1300	8.68E−01	1	0
RNA	CDKN2A	CDKN2A	1029	0.0478	0.0820	8.68E−01	1	0
RNA	PDGFRB	PDGFRB	5159	−0.0355	0.0090	8.71E−01	1	0
RNA	A2M	A2M	2	0.0450	0.0200	8.71E−01	1	0
RNA	ERP29	ERP29	10961	0.0300	−0.0300	8.71E−01	1	0
RNA	PEX11B	PEX11B	8799	0.0360	0.0110	8.71E−01	1	0
RNA	FPGS	FPGS	2356	0.0035	0.0300	8.72E−01	1	0
RNA	PCA3	PCA3	50652	0.1593	0.1240	8.74E−01	1	0
RNA	TUFT1	TUFT1	7286	0.0180	0.0060	8.74E−01	1	0
RNA	CDH11	CDH11	1009	−0.0440	0.0520	8.74E−01	1	0
RNA	OLFML2B	OLFML2B	25903	−0.0102	0.0680	8.76E−01	1	0
RNA	DVL2	DVL2	1856	−0.0580	0.0085	8.77E−01	1	0
RNA	AKR1C3	AKR1C3	8644	0.0068	−0.0490	8.77E−01	1	0
RNA	PRSS8	PRSS8	5652	−0.0075	0.0150	8.80E−01	1	0
RNA	RPEL1	RPEL1	729020	0.0570	0.0090	8.81E−01	1	0
RNA	DYRK4	DYRK4	8798	0.0080	0.0365	8.82E−01	1	0
RNA	CLIC4	CLIC4	25932	0.0100	−0.0450	8.83E−01	1	0
RNA	WDR82	WDR82	80335	−0.0190	−0.0250	8.85E−01	1	0
RNA	STAT3	STAT3	6774	−0.0100	−0.0200	8.86E−01	1	0
RNA	DVL1	DVL1	1855	0.0285	−0.0550	8.87E−01	1	0
RNA	CAPN5	CAPN5	726	0.0873	−0.0330	8.88E−01	1	0
RNA	SPP1	SPP1	6696	0.1305	−0.0740	8.89E−01	1	0
RNA	MPP7	MPP7	143098	−0.0282	−0.0180	8.90E−01	1	0
RNA	RGMB	RGMB	285704	−0.0010	0.0180	8.91E−01	1	0
RNA	ANKRD6	ANKRD6	22881	−0.0030	−0.0410	8.91E−01	1	0
CN (MLPA)	CHD1	CHD1	1105	0.0314	0.0514	8.93E−01	1	1
RNA	WNT11	WNT11	7481	−0.0070	0.0160	8.94E−01	1	0
RNA	CHD1	CHD1	1105	−0.0052	0.0390	8.94E−01	1	0
RNA	PCNA	PCNA	5111	0.0130	0.0160	8.96E−01	1	0
RNA	CYP17A1	CYP17A1	1586	0.0595	−0.0645	8.97E−01	1	0
RNA	COL16A1	COL16A1	1307	0.0340	−0.0300	8.97E−01	1	0
RNA	AGT	AGT	183	0.0115	−0.0520	9.00E−01	1	0
RNA	PDE4DIP	PDE4DIP	9659	−0.0185	−0.0160	9.00E−01	1	0
CN (NanoString)	sig1	NA	NA	0.0949	0.1012	9.00E−01	0	0
RNA	C16orf70	C16orf70	80262	−0.0508	−0.0110	9.02E−01	1	0
RNA	KCNK2	KCNK2	3776	0.0140	0.0440	9.05E−01	1	0
RNA	FBN1	FBN1	2200	0.0325	−0.0020	9.06E−01	1	0
RNA	BPTF	BPTF	2186	0.0290	−0.0070	9.07E−01	1	0
RNA	COL5A1	COL5A1	1289	0.0537	−0.0180	9.07E−01	1	0
RNA	ECU	ECU	1632	−0.0115	−0.0020	9.08E−01	1	0
RNA	GNB4	GNB4	59345	0.0115	−0.0030	9.09E−01	1	0
RNA	HDAC9	HDAC9	9734	−0.0125	0.0295	9.10E−01	1	0
RNA	C9orf152	C9orf152	401546	−0.0415	−0.0500	9.12E−01	1	0
RNA	F3	F3	2152	−0.0600	0.0930	9.17E−01	1	0
RNA	AKT2	AKT2	208	−0.0145	−0.0140	9.18E−01	1	0
RNA	FN1	FN1	2335	0.0397	0.0420	9.19E−01	1	0
RNA	POU5F1	POU5F1	5460	−0.0020	−0.0280	9.26E−01	1	0
RNA	ERG	ERG	2078	0.2938	−0.0270	9.27E−01	1	0
RNA	SIGMAR1	SIGMAR1	10280	0.0105	−0.0020	9.29E−01	1	0
RNA	NFIB	NFIB	4781	0.0197	0.0375	9.30E−01	1	0
RNA	CALM3	CALM3	808	0.0070	0.0030	9.33E−01	1	0
CN (MLPA)	RB1	RB1	5925	0.0238	0.0547	9.34E−01	1	1
CN (MLPA)	GTF2H2	GTF2H2	2966	0.0111	0.0754	9.45E−01	1	1
CN (NanoString)	sig34	NA	NA	0.0569	0.0972	9.83E−01	0	0
CN (MLPA)	CDKN1B	CDKN1B	1027	−0.0237	−0.0108	1.00E+00	1	1
CN (MLPA)	PDPK1	PDPK1	5170	0.0076	−0.0083	1.00E+00	1	1
CN (MLPA)	PDZD2	PDZD2	23037	−0.0041	0.0000	1.00E+00	0	1
CN (MLPA)	RWDD3	RWDD3	25950	0.0112	−0.0045	1.00E+00	1	1
CN (MLPA)	TP53	TP53	7157	0.0490	0.0232	1.00E+00	1	1
CN (NanoString)	sig3	NA	NA	0.0264	0.0360	1.00E+00	0	0
CN (NanoString)	CDKN1B	CDKN1B	1027	0.0073	−0.0026	1.00E+00	0	0
CN (NanoString)	CHD1	CHD1	1105	0.0065	0.0478	1.00E+00	0	0
CN (NanoString)	MYCL1	MYCL	4610	0.0045	−0.0084	1.00E+00	0	0
CN (NanoString)	NKX3-1	NKX3-1	4824	0.0767	0.0641	1.00E+00	0	0
CN (NanoString)	PTEN	PTEN	5728	0.0271	0.1101	1.00E+00	0	0
CN (NanoString)	RB1	RB1	5925	0.0238	0.0675	1.00E+00	0	0
CN (NanoString)	TP53	TP53	7157	0.0800	0.0063	1.00E+00	0	0
CN (NanoString)	sig4	NA	NA	0.0355	0.0410	1.00E+00	0	0
CN (NanoString)	sig5	NA	NA	0.0833	0.0607	1.00E+00	0	0
CN (NanoString)	sig6	NA	NA	−0.0025	0.0039	1.00E+00	0	0
CN (NanoString)	sig7	NA	NA	0.0000	0.0000	1.00E+00	0	0
CN (NanoString)	sig8	NA	NA	0.0025	0.0000	1.00E+00	0	0
CN (NanoString)	sig9	NA	NA	0.0268	0.0000	1.00E+00	0	0
CN (NanoString)	sig10	NA	NA	−0.0215	−0.0168	1.00E+00	0	0
CN (NanoString)	sig11	NA	NA	0.0082	0.0118	1.00E+00	0	0
CN (NanoString)	sig12	NA	NA	0.0503	0.0641	1.00E+00	0	0
CN (NanoString)	sig13	NA	NA	0.0049	0.0000	1.00E+00	0	0
CN (NanoString)	sig14	NA	NA	−0.0463	0.0118	1.00E+00	0	0
CN (NanoString)	sig17	NA	NA	−0.0033	0.0000	1.00E+00	0	0
CN (NanoString)	sig18	NA	NA	−0.0141	0.0000	1.00E+00	0	0
CN (NanoString)	sig19	NA	NA	−0.0083	−0.0045	1.00E+00	0	0
CN (NanoString)	sig20	NA	NA	−0.0116	−0.0168	1.00E+00	0	0
CN (NanoString)	sig23	NA	NA	−0.0043	0.0681	1.00E+00	0	0
CN (NanoString)	sig24	NA	NA	0.0387	0.0612	1.00E+00	0	0
CN (NanoString)	sig25	NA	NA	0.0388	0.0039	1.00E+00	0	0
CN (NanoString)	sig26	NA	NA	−0.0091	−0.0129	1.00E+00	0	0
CN (NanoString)	sig28	NA	NA	0.0007	−0.0129	1.00E+00	0	0
CN (NanoString)	sig29	NA	NA	0.0305	0.0775	1.00E+00	0	0
CN (NanoString)	sig30	NA	NA	0.0083	0.0123	1.00E+00	0	0
CN (NanoString)	sig31	NA	NA	0.0033	0.0000	1.00E+00	0	0
CN (NanoString)	sig32	NA	NA	0.0173	0.0118	1.00E+00	0	0
CN (NanoString)	sig33	NA	NA	0.0223	0.0000	1.00E+00	0	0
CN (NanoString)	sig35	NA	NA	−0.0008	0.0000	1.00E+00	0	0
CN (NanoString)	sig36	NA	NA	0.0165	0.0000	1.00E+00	0	0
CN (NanoString)	sig37	NA	NA	0.0116	0.0163	1.00E+00	0	0
CN (NanoString)	sig39	NA	NA	0.0453	−0.0005	1.00E+00	0	0
CN (NanoString)	sig41	NA	NA	−0.0001	−0.0089	1.00E+00	0	0
CN (NanoString)	sig42	NA	NA	−0.0043	0.0163	1.00E+00	0	0
CN (NanoString)	sig43	NA	NA	−0.0124	0.0247	1.00E+00	0	0
CN (NanoString)	sig44	NA	NA	0.0734	0.0449	1.00E+00	0	0
CN (NanoString)	sig45	NA	NA	0.0032	−0.0134	1.00E+00	0	0
CN (NanoString)	sig46	NA	NA	0.0172	0.0074	1.00E+00	0	0
CN (NanoString)	sig47	NA	NA	0.0568	0.0271	1.00E+00	0	0
CN (NanoString)	sig48	NA	NA	0.0264	−0.0129	1.00E+00	0	0
CN (NanoString)	sig49	NA	NA	−0.0026	0.0163	1.00E+00	0	0
CN (NanoString)	sig50	NA	NA	0.0181	0.0079	1.00E+00	0	0
CN (NanoString)	sig51	NA	NA	0.0297	0.0039	1.00E+00	0	0
CN (NanoString)	sig52	NA	NA	0.0437	0.0528	1.00E+00	0	0
CN (NanoString)	sig53	NA	NA	0.0458	0.0405	1.00E+00	0	0
CN (NanoString)	sig54	NA	NA	0.0578	0.0365	1.00E+00	0	0
CN (NanoString)	sig55	NA	NA	0.0198	0.0123	1.00E+00	0	0
CN (NanoString)	sig56	NA	NA	0.0247	0.1056	1.00E+00	0	0
CN (NanoString)	sig57	NA	NA	0.0049	0.0242	1.00E+00	0	0
CN (NanoString)	sig59	NA	NA	0.0190	0.0039	1.00E+00	0	0
CN (NanoString)	sig60	NA	NA	0.0256	−0.0084	1.00E+00	0	0
CN (NanoString)	sig61	NA	NA	−0.0059	−0.0045	1.00E+00	0	0
CN (NanoString)	sig62	NA	NA	0.0644	0.0730	1.00E+00	0	0
CN (NanoString)	sig63	NA	NA	0.0058	0.0123	1.00E+00	0	0
CN (NanoString)	sig64	NA	NA	0.0058	0.0000	1.00E+00	0	0
CN (NanoString)	sig65	NA	NA	0.0083	0.0000	1.00E+00	0	0
CN (NanoString)	sig66	NA	NA	0.0164	−0.0005	1.00E+00	0	0
CN (NanoString)	sig67	NA	NA	0.0057	−0.0168	1.00E+00	0	0
CN (NanoString)	sig69	NA	NA	0.0255	0.0091	1.00E+00	0	0
CN (NanoString)	sig70	NA	NA	0.0899	0.0458	1.00E+00	0	0
CN (NanoString)	sig71	NA	NA	0.0083	0.0123	1.00E+00	0	0
CN (NanoString)	sig72	NA	NA	−0.0141	0.0079	1.00E+00	0	0
CN (NanoString)	sig74	NA	NA	0.0194	0.0000	1.00E+00	0	0
CN (NanoString)	sig77	NA	NA	0.0033	0.0000	1.00E+00	0	0
CN (NanoString)	sig78	NA	NA	0.0424	0.0242	1.00E+00	0	0
CN (NanoString)	sig79	NA	NA	0.0535	0.0074	1.00E+00	0	0
CN (NanoString)	sig80	NA	NA	0.0083	0.0000	1.00E+00	0	0
CN (NanoString)	sig82	NA	NA	0.0594	0.0562	1.00E+00	0	0
CN (NanoString)	sig83	NA	NA	0.0016	0.0000	1.00E+00	0	0
CN (NanoString)	sig84	NA	NA	−0.0066	0.0000	1.00E+00	0	0
CN (NanoString)	sig85	NA	NA	0.0058	0.0000	1.00E+00	0	0
CN (NanoString)	sig86	NA	NA	0.0058	0.0000	1.00E+00	0	0
CN (NanoString)	sig87	NA	NA	0.0083	0.0039	1.00E+00	0	0
CN (NanoString)	sig88	NA	NA	0.0313	−0.0045	1.00E+00	0	0
CN (NanoString)	sig89	NA	NA	−0.0150	0.0039	1.00E+00	0	0
CN (NanoString)	sig90	NA	NA	0.0049	−0.0005	1.00E+00	0	0
CN (NanoString)	sig91	NA	NA	0.0256	0.0039	1.00E+00	0	0
CN (NanoString)	sig92	NA	NA	−0.0008	0.0123	1.00E+00	0	0
CN (NanoString)	sig93	NA	NA	−0.0033	0.0123	1.00E+00	0	0
CN (NanoString)	sig94	NA	NA	0.0058	0.0123	1.00E+00	0	0
CN (NanoString)	sig95	NA	NA	0.0865	−0.0100	1.00E+00	0	0
CN (NanoString)	sig96	NA	NA	0.0296	−0.0010	1.00E+00	0	0
CN (NanoString)	sig97	NA	NA	0.0750	0.0074	1.00E+00	0	0
CN (NanoString)	sig100	NA	NA	−0.0091	0.0123	1.00E+00	0	0
	MLPA	NanoString		Entrez	Map
	feature	feature	Symbol	gene ID	location
	CDKN1B	CDKN1B	CDKN1B	1027	12p13.1
	CHD1	CHD1	CHD1	1105	5q15-q21.1
	GABARAPL2	NA	GABARAPL2	11345	16q23.1
	GTF2H2	NA	GTF2H2	2966	5q13.2
	MAP3K7	NA	MAP3K7	6885	6q15
	MYC	NA	MYC	4609	8q24.21
	NKX3-1	NKX3-1	NKX3-1	4824	8p21.2
	PDPK1	NA	PDPK1	5170	16p13.3
	PDZD2	NA	PDZD2	23037	5p13.3
	PTEN	PTEN	PTEN	5728	10q23.31
	RB1	RB1	RB1	5925	13q14.2
	RWDD3	NA	RWDD3	25950	1p21.3
	TP53	TP53	TP53	7157	17p13.1
	WRN	NA	WRN	7486	8p12
	NA	MYCL1	MYCL	4610	1p34.2
	NA	sig1	GFRA2	2675	8p21.3
	NA	sig2	CLDN23	137075	8p23.1
	NA	sig2	MFHAS1	9258	8p23.1
	NA	sig2	ERI1	90459	8p23.1
	NA	sig3	ANKRD22	118932	10q23.31
	NA	sig4	STAMBPL1	57559	10q23.31
	NA	sig4	ACTA2	59	10q23.31
	NA	sig4	FAS	355	10q23.31
	NA	sig5	RNLS	55328	10q23.31
	NA	sig6	TAFA5	25817	22q13.32
	NA	sig7	SLC6A19	340024	5p15.33
	NA	sig7	SLC6A18	348932	5p15.33
	NA	sig7	TERT	7015	5p15.33
	NA	sig8	CLPTM1L	81037	5p15.33
	NA	sig9	SLC6A3	6531	5p15.33
	NA	sig10	NKD2	85409	5p15.33
	NA	sig10	SLC12A7	10723	5p15.33
	NA	sig11	TBC1D22A	25771	22q13.31
	NA	sig12	PPP1R3B	79660	8p23.1
	NA	sig13	BRD9	65980	5p15.33
	NA	sig14	TRIP13	9319	5p15.33
	NA	sig17	PDCD6	10016	5p15.33
	NA	sig17	AHRR	57491	5p15.33
	NA	sig17	EXOC3-AS1	116349	5p15.33
	NA	sig18	EXOC3	11336	5p15.33
	NA	sig19	SLC9A3	6550	5p15.33
	NA	sig20	CEP72	55722	5p15.33
	NA	sig20	TPPP	11076	5p15.33
	NA	sig23	LONRF1	91694	8p23.1
	NA	sig24	LIPJ	142910	10q23.31
	NA	sig24	LIPF	8513	10q23.31
	NA	sig24	LIPN	643418	10q23.31
	NA	sig25	CH25H	9023	10q23.31
	NA	sig26	ZBED4	9889	22q13.33
	NA	sig28	ALG12	79087	22q13.33
	NA	sig28	PIM3	415116	22q13.33
	NA	sig28	MLC1	23209	22q13.33
	NA	sig29	TNKS	8658	8p23.1
	NA	sig30	LSM14B	149986	20q13.33
	NA	sig30	SS18L1	26039	20q13.33
	NA	sig30	MTG2	26164	20q13.33
	NA	sig31	PTPRT	11122	20q12-q13.11
	NA	sig32	ANKRD10	55608	13q34
	NA	sig33	MGMT	4255	10q26.3
	NA	sig33	EBF3	253738	10q26.3
	NA	sig33	GLRX3	10539	10q26.3
	NA	sig34	CTSB	1508	8p23.1
	NA	sig34	DEFB134	613211	8p23.1
	NA	sig35	PREXI	57580	20q13.13
	NA	sig36	CCDC127	133957	5p15.33
	NA	sig36	SDHA	6389	5p15.33
	NA	sig37	ATP11AUN	400165	13q34
	NA	sig37	ATP11A	23250	13q34
	NA	sig37	MCF2L	23263	13q34
	NA	sig39	SSPO	23145	7q36.1
	NA	sig41	BECN1	8678	17q21.31
	NA	sig41	PSME3	10197	17q21.31
	NA	sig41	AOC3	8639	17q21.31
	NA	sig42	ZNF862	643641	7q36.1
	NA	sig43	ATP6V0E2	155066	7q36.1
	NA	sig44	CHRNA6	8973	8p11.21
	NA	sig45	KDM6B	23135	17p13.1
	NA	sig45	CHD3	1107	17p13.1
	NA	sig46	GUCY2D	3000	17p13.1
	NA	sig47	ALOX15B	247	17p13.1
	NA	sig48	ALOX12B	242	17p13.1
	NA	sig49	PER1	5187	17p13.1
	NA	sig49	AURKB	9212	17p13.1
	NA	sig50	PFAS	5198	17p13.1
	NA	sig50	SLC25A35	399512	17p13.1
	NA	sig50	RANGRF	29098	17p13.1
	NA	sig51	SMIM19	114926	8p11.21
	NA	sig52	POLB	5423	8p11.21
	NA	sig52	DKK4	27121	8p11.21
	NA	sig53	VDAC3	7419	8p11.21
	NA	sig53	SLC20A2	6575	8p11.21
	NA	sig54	AP3M2	10947	8p11.21
	NA	sig54	PLAT	5327	8p11.21
	NA	sig55	IL9	3578	5q31.1
	NA	sig55	FBXL21P	26223	5q31.1
	NA	sig55	LECT2	3950	5q31.1
	NA	sig56	MTMR9	66036	8p23.1
	NA	sig57	HTR3A	3359	11q23.2
	NA	sig57	NNMT	4837	11q23.2
	NA	sig59	BUD13	84811	11q23.3
	NA	sig59	ZPR1	8882	11q23.3
	NA	sig59	APOA1	335	11q23.3
	NA	sig60	RNF214	257160	11q23.3
	NA	sig60	CEP164	22897	11q23.3
	NA	sig61	FXYD6	53826	11q23.3
	NA	sig62	GATA4	2626	8p23.1
	NA	sig62	NEIL2	252969	8p23.1
	NA	sig62	FDFT1	2222	8p23.1
	NA	sig63	ZNF618	114991	9q32
	NA	sig63	KIF12	113220	9q32
	NA	sig63	COL27A1	85301	9q32
	NA	sig64	ZNF334	55713	20q13.12
	NA	sig64	TP53RK	112858	20q13.12
	NA	sig64	EYA2	2139	20q13.12
	NA	sig65	NCOA3	8202	20q13.12
	NA	sig66	SULF2	55959	20q13.12
	NA	sig67	MAFB	9935	20q12
	NA	sig69	ZMYND11	10771	10p15.3
	NA	sig70	DIP2C	22982	10p15.3
	NA	sig71	IDI2	91734	10p15.3
	NA	sig71	WDR37	22884	10p15.3
	NA	sig72	ADARB2	105	10p15.3
	NA	sig74	LPCAT1	79888	5p15.33
	NA	sig77	TAF4	6874	20q13.33
	NA	sig7S	SLC7A5	8140	16q24.2
	NA	sig7S	CA5A	763	16q24.2
	NA	sig79	NRG3	10718	10q23.1
	NA	sig80	TOP1	7150	20q12
	NA	sig80	ZHX3	23051	20q12
	NA	sig80	CHD6	84181	20q12
	NA	sig82	SGK2	10110	20q13.12
	NA	sig83	IFT52	51098	20q13.12
	NA	sig83	MYBL2	4605	20q13.12
	NA	sig84	GTSF1L	149699	20q13.12
	NA	sig84	TOX2	84969	20q13.12
	NA	sig85	NCOA5	57727	20q13.12
	NA	sig85	CD40	958	20q13.12
	NA	sig85	SLC35C2	51006	20q13.12
	NA	sig86	ARFGEF2	10564	20q13.13
	NA	sig87	MOV10L1	54456	22q13.33
	NA	sig88	PANX2	56666	22q13.33
	NA	sig89	HDAC10	83933	22q13.33
	NA	sig89	PLXNB2	23654	22q13.33
	NA	sig90	MIOX	55586	22q13.33
	NA	sig90	CPT1B	1375	22q13.33
	NA	sig90	MAPK8IP2	23542	22q13.33
	NA	sig91	RAB22A	57403	20q13.32
	NA	sig91	APCDD1L	164284	20q13.32
	NA	sig91	NPEPL1	79716	20q13.32
	NA	sig92	OSBPL2	9885	20q13.33
	NA	sig92	RPS21	6227	20q13.33
	NA	sig92	SLCO4A1	28231	20q13.33
	NA	sig93	OGFR	11054	20q13.33
	NA	sig93	TCFL5	10732	20q13.33
	NA	sig94	GNAS	2778	20q13.32
	NA	sig94	TUBB1	81027	20q13.32
	NA	sig94	PRELID3B	51012	20q13.32
	NA	sig95	ZFPM1	161882	16q24.2
	NA	sig96	ZC3H18	124245	16q24.2
	NA	sig96	CYBA	1535	16q24.2
	NA	sig96	MVD	4597	16q24.2
	NA	sig97	SNAI3	333929	16q24.2
	NA	sig97	RNF166	115992	16q24.2-q24.3
	NA	sig100	TPRG1L	127262	1p36.32
	NA	sig100	TP73	7161	1p36.32
	NA	sig100	CCDC27	148870	1p36.32

Claims

1. A method of predicting disease progression risk in a subject with prostate cancer, the method comprising: a) providing a sample containing RNA and DNA material from tumour cells;b) determining or measuring values for substantially all of 353 patient features comprising the mRNA and copy number aberration (CNA) features listed for PRONTO-e in Table 6, and some or all reference or control features set forth in Table 6;c) comparing said patient features to the reference or control features; andd) computing a prediction score using a classifier that takes said patient feature values as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

2. The method of claim 1, wherein substantially all of 353 patient features is all 353 patient features.

3. The method of claim 1, wherein determining the prediction score comprises classifying the patient tumour into a pathological Gleason Grade Group (GG) class.

4. The method of claim 1, wherein the patient tumour is classified in the pathologic GG≥2 class if the score is ≥0.5 or the pathologic GG1 class if the score is <0.5.

5. The method of claim 3, wherein if the patient is classified into the pathologic GG1 class, further comprising managing the patient with active surveillance.

6. The method of claim 3; wherein if the patient is classified into the pathologic GG≥2 class, further comprising treating the patient with surgery, endocrine therapy, chemotherapy, radiotherapy, hormone therapy, gene therapy, thermal therapy, or ultrasound therapy.

7. A method of predicting disease progression risk in a subject with prostate cancer, the method comprising: a) providing a sample containing RNA and DNA material from tumour cells;b) determining or measuring substantially all of 94 patient features comprising the mRNA, CNA, methylation and clinical features listed for PRONTO-m in Table 6, and some or all reference or control features set forth in Table 6;c) comparing said patient features to the reference or control features; andd) computing a prediction score using a classifier that takes said patient feature values as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

8. The method of claim 7, wherein substantially all of 94 patient biomarkers is all 94 patient biomarkers.

9. The method of claim 7, wherein determining the prediction score comprises classifying the patient tumour into a pathological Gleason Grade Group (GG) class.

10. The method of claim 7, wherein the patient tumour is classified in the pathologic GG≥2 class if the score is ≥0.5 or the pathologic GG1 class if the score is <0.5.

11. The method of claim 9, wherein if the patient is classified into the pathologic GG1 class, further comprising managing the patient with active surveillance.

12. The method claim 9, wherein if the patient is classified into the pathologic GG≥2 class, further comprising treating the patient with surgery, endocrine therapy, chemotherapy, radiotherapy, hormone therapy, gene therapy, thermal therapy, or ultrasound therapy.

13. A computer-implemented method of predicting disease progression risk in a patient with prostate cancer, the method comprising: a) receiving, at at least one processor, data reflecting substantially all of the patient features defined in claim 1 corresponding to the PRONTO-e or PRONTO-m classifiers regarding a prostate cancer tumor, and some or all reference or control features set forth in Table 6;b) constructing, at at least one processor, a patient profile based on the patient features;c) comparing, at the at least one processor, said patient profile to the reference or control;d) computing, at the at least one processor, a prediction score using a classifier that takes said patient profile as input, the classifier having been previously trained on samples from a population of early prostate cancer patients.

14. The method of claim 13, wherein substantially all patient features is all 353 patient features in the case of PRONTO-e and all 94 patient features in the case of PRONTO-m.

15. The method of claim 13, wherein computing the prediction score comprises classifying the patient tumour into a pathological GG class.

16-20. (canceled)