Ant Colony Optimization for Prediction of Compound-Protein Interactions

Abstract

The prediction of Compound-Protein Interactions (CPI) is an essential step in drug-target analysis for developing new drugs. Therefore, it needs a good incentive to develop a faster and more effective method to predicting the interaction between compound and protein. Predicting the unobserved link of CPI can be done with Ant Colony Optimization for Link Prediction (ACO_LP) algorithms. Each ant selects its path according to the pheromone value and the heuristic information in the link. The path passed by the ant is evaluated and the pheromone information on each link is updated according to the quality of the path. The pheromones on each link are used as the final value of similarity between nodes. The ACO_LP are tested on benchmark CPI data: Nuclear Receptor, G-Protein Coupled Receptor (GPCR), Ion Channel, and Enzyme. Result show that the accuracy values for Nuclear Receptor, GPCR, Ion Channel, and Enzyme dataset are 0.62, 0.62, 0.74, and 0.79 respectively. The results indicate that ACO_LP has good accuracy for prediction of CPI.