Affinity Prediction Models from Protein-Protein Interaction of SCA Using Ensemble Learning

Abstract

Remedy of inherited disorder like spinocerebellar ataxia (SCA) is a challenge and a necessary task in biomedical research. There are number of approaches available for affinity prediction through various scores and features in a standard computational framework hence it is significant to depict the binding affinity for drug identification. Affinity prediction is incredibly significant for drug discovery and it involves numerous steps like active site identification and docking. Identification of active site is obligatory for proteins to interact either with ligands or protein.  The main focus of this work is to utilize the ensemble learning methods to build model for affinity prediction through 3d protein structures and interactive properties of that structures which can be used for predicting binding affinity. Protein-protein interaction is performed and the binding affinity is calculated from the interacted complex. Features like physio-chemical properties, energy calculations, interfacial and non-interfacial properties are extracted from the interacted complexes to construct enhanced predictions. Ensemble learning scheme is meta algorithms that coalese numerous machine learning techniques into one predictive model in order to lessen variance, bias, or improve predictions. The Random forest regressor in forest of randomized trees performs better by combining many algorithms. Experiments discovered the dominance of random forest regressor in forest of randomized trees when compared to other ensemble learning methods.