Evolving relational hierarchical classification rules for predicting gene ontology-based protein functions

Abstract

Hierarchical Multi-Label Classification (HMC) is a complex classification problem where instances can be classified into many classes simultaneously, and these classes are organized in a hierarchical structure, having subclasses and superclasses. In this paper, we investigate the HMC problem of assign functions to proteins, being each function represented by a class (term) in the Gene Ontology (GO) taxonomy. It is a very difficult task, since the GO taxonomy has thousands of classes. We propose a Genetic Algorithm (GA) to generate HMC rules able to classify a given protein in a set of GO terms, respecting the hierarchical constraints imposed by the GO taxonomy. The proposed GA evolves rules with propositional and relational tests. Experiments using ten protein function datasets showed the potential of the method when compared to other literature methods.