As the Human Genome Project nears completion the need grows for functional genomic analyses which in addition to the primary genomic sequence involve other types of data such as gene expression measurements from microarray hybridization experiments. Research in functional genomics involves a range of computational problems including visualization, clustering, classification, regression, knowledge representation, and predictive modeling. This project will touch on each of these areas, but the primary focus will be on developing machine learning techniques that learn to place genes into discrete functional categories in order to simplify and render more tractable the problem of inferring gene function from genomic data. To the end the PI will build on his prior work which showed that a support vector machine (SVM) can be successfully trained using DNA microarray expression data to recognize various gene functional categories, and will develop methods for combining coding sequence, promoter region, gene expression, and other types of genomic data in SVM-based learning algorithms. The research will lead to improved understanding of the ability of various machine learning techniques to recognize different types of gene functional classes, and will also yield new techniques for learning simultaneously from multiple types of data. Learning from heterogeneous data sets is a core issue in artificial intelligence and machine learning; the ability to combine knowledge from various types of genomic data is critical for understanding the cell at the molecular level, and should lead to important insights into gene function.
