Improving an evolutionary multi-objective algorithm for the biclustering of gene expression data

The development of new technologies for the design of DNA microarrays has boosted the generation of large volumes of biological data, which requires the development of efficient computational methods for their analysis and annotation. Among these methods, biclusters generation algorithms attempt to identify coherent associations of genes and experimental conditions. In this paper, we introduce an improved version of a multi-objective genetic algorithm to find large biclusters that are, at the same time, highly homogeneous. The proposed improvement uses a group based representation for the genes-conditions associations rather than long binary strings. To assess the proposal performance the algorithm is applied to generate biclusters for two real gene expression data: Saccharomyces Cerevisiae with 2884 genes and 17 conditions, and the human B cells Lymphoma with 4026 genes and 96 conditions. The results of computational experiments show that the proposed approach outperforms current state-of-the-art algorithms on these data sets.