FEATURIZATION OF DRUG COMPOUNDS AND TARGET PROTEINS FOR DRUG-TARGET INTERACTION PREDICTION

Abstract

In recent times, computational methods for Drug-Target Interaction (DTI) prediction have become pervasive due to the advances in computational resources and techniques. Computational methods, particularly Machine Learning (ML) based methods, are very efficient DTI predictors. They help greatly in reducing the search space of drug candidates, consequently reducing the cost and time for drug development. Featurization is an important task in machine learning-based DTI prediction methods. Descriptors of drug compounds and target proteins are computed with available Cheminformatics software packages and fed as inputs to ML models for DTI prediction. The process of featurization poses a challenge to non-experts with little or no knowledge of molecular fingerprints, descriptors, and computational toolkits in the domain of Cheminformatics. In this study, we aim to provide reference and tutorial insights into featurization of drug compounds and proteins for DTI prediction. Firstly, we present an overview of DTI prediction encompassing definitions of key terms, various categories of computational methods, and bioactivity repositories. Subsequently, we discuss feature extraction, drug compound descriptors, protein descriptors, and Cheminformatics toolkits. There is a demonstration of how Cheminformatics toolkits can be used for featurization and a review of recent works that constructed compound and protein feature vectors for DTI prediction. Lastly, we give a summary of the study and highlight the challenges of featurization. This is the maiden work that discusses featurization of compounds and proteins for DTI prediction and gives practical examples of how it is done.