NSF Award
0523643
Progress in the biological sciences in the post-genomic era depends on our ability to make sense of genome-scale information. The genome and the proteome together establish an intricate network of interactions that exhibits many similarities to social and political networks. This interaction network forms a simple, conceptual representation of the molecular machinery in the living cell. Knowledge of individual interactions and patterns of interaction therefore significantly enhances our understanding of the mechanisms of biological function. <br/><br/>Unfortunately, experimental determination of molecular interactions at the scale of the entire genome is often error-prone: many interactions revealed by such high-throughput experiments are false, and conversely, many of the actual interactions are not revealed at all. It is important, therefore, to establish rigorous computational methods that utilize high-throughput data from a variety of sources to predict the existence (and lack thereof) of an interaction, and to assign confidence levels to each prediction in a systematic manner. Our project utilizes state-of-the-art predictive methods from the field of Artificial Intelligence-Bayesian support vector machines-to predict molecular interactions at the whole-genome level. The project is initiated by an exhaustive data collection effort involving a variety of data sources that supply putative predictors for the presence or absence of interactions among protein/protein or gene/protein pairs. Dominant predictors among these will be isolated, and the prediction system will be applied to the genomes of several organisms, including the budding yeast, worm, and fly. The accuracy of the method will be tested and refined by computational and biological means. Successful completion of this project will significantly enhance our ability to decipher genomic information and apply these findings to discover novel functional pathways of biological, agricultural, and medical importance. <br/><br/>All methods and results will be publicly disseminated, the former with stand-alone executable programs, and the latter via publications and web pages. The project will support interdisciplinary training of graduate and postdoctoral students, and should provide research opportunities for undergraduate students.
