Extracting Explicable Rules for the Identification of Compound–Protein Interactions

Abstract

Compound-protein interaction (CPI) is one of the essential interaction patterns in living organisms. However, its underlying mechanism has not been fully revealed because of its complicated processes. Determining CPIs with traditional experiments can reveal solid results. However, their defects, such as low efficiency and high cost, are also evident. Designing effective computational methods is an alternative way to determine CPIs. Several methods have been proposed, but such methods can provide limited information to reveal the mechanism of CPIs because most of them are black boxes. In this study, we tried to develop rule-based classifiers for the identification of CPIs. The obtained rules involved gene ontology, KEGG pathway, and molecular ACCess System fingerprint descriptor, which could describe the functional enrichment of CPIs, to constitute the criterion. Although the performance of rule-based classifiers was lower than that of previous black box classifiers, these classifiers could clarify the identification procedures and provide more information on the mechanism of CPIs. The reliability of the obtained rules was also analyzed.

Keywords

• Compound–protein interaction
• gene ontology
• KEGG pathway
• molecular fragment
• explicable rule
• RIPPER