Multiobjective Differential Evolution for Predicting Protein-Protein Interactions Using GO-Based Semantic Similarity Measures

Semantic similarity is useful to assess the degree of relatedness of two entities by means of their annotations. Numerous biomedical studies have used Gene Ontology (GO)-based semantic similarity to validate their results. To compute the similarity between two interacting proteins, many semantic similarity measures are available under the three GO categories: biological process (BP), molecular function (MF), and cellular component (CC). This study aims to identify a set of semantic similarity measures that are useful for predicting protein-protein interactions. For this purpose, a Differential Evolution for Multiobjective Optimization (DEMO)-based feature selection approach is developed. It identifies a subset of semantic similarity measures that can predict protein-protein interactions while optimizing multiple classification metrics at the same time. The efficacy of the proposed multiobjective technique is compared with some existing feature selection methods. The proposed multiobjective technique has been found to outperform the other methods in most cases.