Application of machine leaning approaches in drug target identification and network pharmacology

Drugs may interact with numerous molecules in the human body. Unexpected drug-protein binding (e.g. off-target binding interactions) may result in adverse reactions, which increase therapeutic risks and negatively impact drug development. Applying network pharmacology to predict drug effects and toxicity resulting from multi-target interactions is therefore considered a promising method capable of comprehensively assessing pharmacological effects. Here, we present a novel screening approach which combines two elaborately developed machine learning systems and the use of multiple docking packages to assess binding potentials of a test compound against proteins involved in a complex molecular network. We developed two machine learning systems, including a re-scoring function to assess binding modes generated by docking tools and to rank them accordingly, and a binding mode selection function designed to identify the most predictive binding mode. The developed machine learning systems significantly enhance the prediction accuracy. We applied these two learning systems to screen a number of test compounds over Influenza A Virus Life Cycle pathway map (FluMap) for discovering anti-influenza agents. Using the method, we have screened the compounds targeting the host proteins identified by siRNA screening, and identified those showing efficacy in vitro. Together with the application of network pharmacology, the proposed screening approach is able to comprehensively characterize the underlying mechanism of a drug candidate with good prediction accuracy. This would contribute in reducing the number of tests for further bioassay and lead a step change in the prediction of drug efficacy and safety.)