Drug Development for Hepatitis C Virus Infection: Machine Learning Applications

Abstract

Hepatitis C virus (HCV) infection is one of the leading causes of mortality and morbidity, and is widely reported for its association with the development of liver cirrhosis, hepatocellular cancer, and liver failure. Most of the reported cases of hepatitis C end up with a chronic form of the infection, existing as a large threat for public health and can be prevented by evading or eradicating the virus through effective drug development. Conventional medicines that are both safe and easily affordable, have not yet been developed for the treatment of chronic HCV infection. Apart from only identifying novel drugs, it is equally important to explore their effectiveness by ascertaining drug target accuracy, which is a crucial part of any drug development program. Moreover, it is highly critical to understand the activity and molecular basis of drug resistance of various drugs, as they may retain activity against a broad spectrum of drug resistant viral variants. Drug discovery and design are highly complex, time consuming, and expensive endeavors. Therefore, it is crucial to incorporate new technologies for this process. Modern drug design strategies include ligand-based (LBDD) and structure-based drug design (SBDD) methods to develop new drug candidates. Machine Learning (ML) approaches are extensively applied in drug design processes for HCV and most common applications include classifying drug targets into druggable and non-druggable, prioritizing drug targets, discovering novel inhibitors, predicting diseases by using risk factors as classifiers, in silico ADMET prediction, etc. However, a few studies using Machine Learning approaches have been reported for prediction of biological activity from multivariate models, prediction of binding site secondary structural modes of docking, and virtual screening.

The most common ML techniques applied in HCV drug discovery, comprise techniques such as random forest, SVM, Decision tree, Genetic algorithms, K-Nearest Neighbor’s, Naive Bayesian classifiers, Particle swarm optimization, as well as multilinear regression models. These tools are widely used in drug discovery studies as they are readily accessible, both as open source and commercial distributions, statistically consistent, computationally efficient, and relatively straight-forward to implement and interpret. Moreover, data-mining software enables users to implement these algorithms through graphical user interfaces and can also be written and executed using packages such as R, Matlab, and Octave. Datamining and Machine Learning approaches hence seem as promising aid for Drug Development studies on HCV infection.