Ranking Genes by Relevance to a Disease

The problem of identifying key genes that are involved in a particular disease is of fundamental importance in biology and medicine. Given the increasing availability of a variety of gene-related biological data sources, ranging from microarray expression data to protein-protein interaction data, a promising approach is to use bioinformatics methods that can analyze this data and rank genes based on potential relevance to a disease; such methods can be invaluable in helping to prioritize genes for further biological study. Recently, the problem of ranking objects has gained considerable attention in machine learning and data mining; ranking problems arise in a variety of domains ranging from document retrieval to collaborative filtering, and a variety of new learning methods have been developed that directly optimize ranking performance. Here we propose for the first time the use of such ranking methods for the problem of ranking genes. We illustrate this approach on microarray data for leukemia and colon cancer; in both cases, the ranking methods we use identify several genes that were not identified by previous methods, including some novel genes that could potentially evolve as targets for drug development. Our study suggests that ranking methods in machine learning could emerge as a powerful tool for mining gene-related data sources for the identification of genes relevant to a particular disease.