Classification and sparse-signature extraction from gene-expression data

Abstract

In this work we suggest a statistical mechanics approach to the classification of highdimensional data according to a binary label. We propose an algorithm whose aim istwofold: first it learns a classifier from a relatively small number of data; second it extractsa sparse signature, i.e., a lower dimensional subspace carrying the information needed forclassification. In particular the second part of the task is NP-hard; therefore we propose astatistical mechanics based message-passing approach. The resulting algorithm is testedon artificial data to prove its validity, but also to elucidate possible limitations.As an important application, we consider the classification of gene-expression datameasured in various types of cancer tissues. We find that, despite the currently lowquantity and quality of available data (the number of available samples is much smallerthan the number of measured genes, thus strongly limiting the predictive capacities), thealgorithm performs slightly better than many state-of-the-art approaches in bioinformatics.