Redescription Mining and Applications in Bioinformatics

Our ability to interrogate the cell and computationally assimilate its answers is improving at a dramatic pace. For instance, the study of even a focused aspect of cellular activity, such as gene action, now benefits from multiple high-throughput data acquisition technologies such as microarrays, genome-wide deletion screens, and RNAi assays. A critical need is the development of algorithms that can bridge, relate, and unify diverse categories of data descriptors. Redescription mining is such an approach. Given a set of biological objects (e.g., genes, proteins) and a collection of descriptors defined over this set, the goal of redescription mining is to use the given descriptors as a vocabulary and find subsets of data that afford multiple definitions. The premise of redescription mining is that subsets that afford multiple definitions are likely to exhibit concerted behavior and are, hence, interesting. We present algorithms for redescription mining based on formal concept analysis and applications of redescription mining to multiple biological datasets. We demonstrate how redescriptions identify conceptual clusters of data using mutually reinforcing features, without explicit training information.