A comparative analysis of Drug-Target Interaction detection approaches using Ensemble of Machine learning models
Simrin Baruah, Debashree Devi · 2024
Drug Discovery is the process of identifying the chemical entities that have the potential to become a medicine. The first and crucial step in drug discovery is to predict a target protein; which is intended to act on the drug-in-development. Hence, recognition of the accurate Drug-Target Interaction (DTI) pair is an important task in the domain of effective drug discovery. The traditional methods of drug discovery are time-consuming and expensive due to the involvement of wet-lab experimentations, and manual chores. Moreover, size of the unknown/ unidentified DTI pairs is higher than known/ identified DTI pairs, which leads to the problem imbalanced data. In this paper, at first, the different methods for generating the features for the drugs and the proteins are investigated, along with a set of feature selection techniques to retrieve the optimal features set from four datasets. Following this, data balancing is performed to resolve the issue of imbalanced known-unknown DTI pair detection. Eight pipelines of different feature generation-feature selection-data balancing-classification strategies are investigated and the empirical analyses provides to determine the most effective pipeline for efficient DTI detection.