This award provides funding for the development of novel optimization-based methods for handling the problem of biclustering in biomedical data analysis. Biclustering consists of simultaneously partitioning the set of data samples and the set of their attributes (features) into subsets (clusters). Biclustering has great significance for data analysis in a variety of biomedical applications. Performing it with high reliability, we are able to not only diagnose conditions represented by sample clusters, but also to identify the key features (e.g., genes) responsible for them, or serving as their markers. The criteria used to relate clusters of samples and clusters of features may differ, relying on various patterns of interest among elements of a bicluster. The search for these patterns can be represented as optimization tasks, the solutions of which form the biclusters of data. We propose a family of new mathematical programming models to tackle the biclustering problem in biomedical data analysis. Advanced solution approaches will be developed for solving the proposed optimization models. Computational experiments involving real data analysis problems will be performed to validate the implemented algorithms against existing methodologies.
If successful, the results of this research project are expected to have a high impact in the area at the intersection of optimization, data mining and biomedicine. It will facilitate the expansion of advanced biclustering analysis in related health care and biomedical applications, potentially improving prediction and diagnosis procedures, as well as the understanding of the mechanisms responsible for many diseases. The proposed work is also expected to enhance the existing computational tools and methodologies for solving hard optimization problems.
