Generating the logicome from microarray data

Abstract

The advances in complex statistics and machine learning methods lead to the development of powerful classifiers that can be used to recognize cellular states (such as gene expression profiles) that are associated to a number of gene-scale expressed diseases, for instance, cancer. However, the data-driven models built by means of learning from datasets in a number of cases represent “black boxes” that cannot be easily analyzed and understood. In this article, we suggest a method for building a data-driven logicome. I.e., the method for building a set of small boolean expressions as classifiers for disjoint groups of samples from a microarray dataset. We validate our method on the microarray dataset of head and neck/oral squamous cell carcinoma, where our boolean signature presented a set of gene activity/inactivity combinations that are characteristic for various cancer sub-types and normal samples. Our findings correlate well with the literature.