Relational Descriptive Analysis of Gene Expression Data

This paper presents a method that uses gene ontologies, together with the paradigm of relational subgroup discovery, to help find description of groups of genes differentially expressed in specific cancers. The descriptions are represented by means of relational features, extracted from publicly available gene ontology information, and are straightforwardly interpretable by the medical experts. We applied the proposed method to two known data sets: (i) acute lymphoblastic leukemia (ALL) vs. acute myeloid leukemia (AML) and (ii) classification of fourteen types of cancer. Significant number of discovered groups of genes had a description, confirmed by the medical expert, which highlighted the underlying biological process that is responsible for distinguishing one class from the other classes. We view our methodology not just as a prototypical example of applying more sophisticated machine learning algorithms to gene expression analysis, but also as a motivation for developing increasingly more sophisticated functional annotations and ontologies, that can be processed by such learning algorithms.