A Hands-On Perspective on Physico-chemical Versus AI/ML Methods Along the Genome to Drug Pathway

The availability of genomic sequences in public domain has spurred intense algorithmic developments for identifying protein-coding regions (gene finding) and for establishing their functions. This in turn accelerated protein structure prediction efforts, particularly relevant for proteins crucial to pathogens and those involved in disease conditions, to further enable structure-based drug design endeavours. This modern drug discovery pathway (genome -> gene -> protein -> drug) in our hands became a set of software suites collectively called ‘Dhanvantari’, which embodies Chemgenome, Bhageerath and Sanjeevini web suites to traverse through the genome to drug pathway with entry along any point. While developing the above science and software suites, our focus has been on energy, forcefield and molecular simulation-based methods, which are referred to here as physico-chemical methods. AI/ML methods have literally stormed in during the last few years into these research areas making it almost difficult to ignore their strengths. The result is the emergence of integrated AI/ML and physico-chemical methods for improved accuracies and greater success rates in new molecule predictions against drug targets. This brief report sketches how these methods have evolved in our hands to help accelerate drug discovery.