Disease Classification through Integer Optimisation

Abstract In microarray data analysis, traditional methods focusing either on all the genes or a single gene at a time are being replaced by methods based on sets of genes that correspond to biochemical pathways, to offer more informative strategies into disease associations. However, the development of robust pipelines to relate the genotype to disease phenotypes through known molecular interactions is still in its early stages. We report the use of a mathematical optimisation approach based on hyper-box principles to classify cancer samples within pathways into appropriate disease phenotypes. Most informative genes were identified based on non-overlapping constraints of the classification procedure and the algorithm showed good performance comparing to established classification protocols.