Research Project
10304324
The Data Science Core (DSC) will provide critical support for the P01 project as a whole to ensure its success by offering a central source related to research design, data management, statistical analysis and machine learning. The DSC has assembled a team of highly qualified investigators with a broad range of expertise in HIV research including design of clinical trials, statistical inference methods, integration of diverse -omics data and neuroimaging data, data management, data security, machine learning/artificial intelligence (ML/AI), and analytics. The DSC will also provide training services in collaboration with the training programs in other components of this P01. In addition to supporting the proposed two intervention studies in the P01, the DSC will leverage existing data resources to test important hypotheses and build prediction models and personalized recommendation tools for treating HIV infections for patients who are heavy drinkers. When the data from Projects 1 and 2 are available, cross-cohort prediction and personalized recommendation tool will be constructed with state-of-the-art statistical learning and machine learning techniques. Specifically, our aim one will provide support in study design, data management, data sharing, statistical analysis, and research dissemination to ensure proper and efficient conduct of the two research projects. Working closely with the Administrative Core and two project teams, this aim will carry out a series of tasks including (but not limited to): development of centralized study database and web-based Electronic Data Capture (EDC) system; generate randomization schemes; design and implement quality control procedures for data collection/processing; train site staff in the use of data collection and data management system; provide support in data masking, data harmonization, and data sharing. Based on the existing data from the Thirty-Day Challenge Study, our aim 2 will perform causal analysis and AI modeling to explore causal relationships between baseline characteristics, changes in alcohol use, changes in neuroimaging and microbiome biomarkers, and changes in neurocognitive functions. This aim will build a baseline prediction model to predict change in alcohol use after the intervention wit baseline information. Multi-scale dynamic modeling will be used to integrate voxel-level, tissue-level, region-level, and lobe-level neuroimaging information for prediction of alcohol abstinence. We will also identify the key changes in multimodal neuroimaging and microbiome biomarkers associated with levels of alcohol abstinence. Direct effects of baseline characteristics on changes in neurocognitive functions, and their indirect effects through changes in alcohol use, neuroimaging and microbiome biomarkers will be estimated and tested. Our aim 3 will use the data from two new randomized clinical trials to validate and refine prediction models developed in Aim 2 and build a personalized intervention recommendation tool. Cross-cohort validation will be conducted in each of the two new clinical trials using established protocols and in the pooled data of the two trials to validate and refine the baseline prediction models for predicting alcohol use reduction. Longitudinal cross-cohort learning will be employed to create a uniform prediction model across three research projects and build a personalized intervention recommendation tool.
