Rare target detection with statistical learning methods: drug discovery, functional genomics and proteomics

点击翻译按钮获取中文摘要

My proposed research will concentrate on an important and relatively new class of learning problems in data mining that I shall refer to as "rare target detection problems." The underlying objective is to detect items belonging to a rare but important class from very large databases. Common applications include drug discovery, fraud detection, direct marketing, as well as problems in the newly emerged disciplines of bioinformatics such as functional genomics and proteomics. For example, in functional genomics, we are often interested in finding a few genes that are associated with a particular biological function; in drug discovery, we are interested in identifying a few chemical compounds (from amongst tens of thousands of candidates) that are capable of inhibiting a certain protein. Accurate and efficient algorithms for solving such problems will have a significant and broad impact on the advancement of science and technology. My objectives are: to better understand the statistical nature of the rare target detection problem, to develop effective and efficient learning algorithms for solving it; to apply these algorithms to solve various real-world rare target detection problems and to promote the use of these and other learning algorithms in applied research. Recently, working with a graduate student and a colleague, I have invented a novel and efficient detection algorithm called LAGO. In developing LAGO,  I have also laid out a systematic framework as a foundation for further research. The research program outlined in this proposal is built on this foundation and aimed at deepening and broadening its impact.