Abstract: The ability to control infections has been continuously threatened by the emergence and spread of drug resistance. As capacity to develop novel drugs hasn’t significantly improved, computationally characterising drug resistance mutations has become paramount for stewardship, pathogen monitoring and to guide drug discovery. Here we describe mmCSM-lig, the first accurate and scalable computational platform capable of quantitatively assessing how single and multiple mutations directly affect protein-ligand binding and its links to potential phenotypes, from drug to herbicide resistance. The method is freely available as and easy-to-use web interface and API, providing programmatic access for integration with analytical pipelines. We demonstrate our method significantly outperforms alternative methods and its previous version, being capable of predicting effects of multiple mutations. mmCSM-lig was trained on hand curated data from the literature, including almost 1,000 single and multiple mutations from over 200 protein-ligand complexes, with experimentally characterised 𝚫𝚫Gs (different in energy of binding in Kcal/mol). The model was internally validated via different cross validation schemes as well as with low-redundancy, independent blind tests, achieving consistent and robust performance, significantly outperforming alternative methods.
mmCSM-Lig is freely available by user-friendly web-interface and API at https://biosig.lab.uq.edu.au/mmcsm_lig/ .

This website is free and open to all users and there is no login requirement.