The RESPOND3 project sits on 4 work packages: 1) Computational drug-target affinity prediction, 2) Medicinal Chemistry, 3) Compound Assays and Structural Biology, and 4) Responsible Research and Innovation. This linear enumeration does not make justice to their interdependence. Packages 1), 2) and 3) are the three pillars of the modern structure-based drug design cycle; they inform each other iteratively: computational predictions of binding affinity allow to prioritize compounds to be synthesized which in turn can be tested for biological activity. What about the work package 4)? It is the medium in which the others are embedded, since responsible research and transparency is of utmost importance for a research project that balances the interests of different stakeholders: researchers, private investors, people in need of the drug, government, society. And the stakes are high for everyone involved when millions are invested, thousands of hours of labor are put forward, and the outcome is uncertain.
The research environment of the project provides the perfect playground to develop better computational tools to aid drug discovery. The reason is that the methods can not only be tested retrospectively, but also prospectively within the project, for example in aiding the development of therapeutics for Chronic Obstructive Pulmonary Disease (COPD).
This live feedback is key for the computational tools developers to bring about a tool that is useful for the medicinal chemists. Thanks to the growing amount of relevant drug discovery data available in public databases the use of modern machine learning methods is being explored to develop approaches that can further streamline drug discovery. The use of machine learning presents the potential to curb the computational cost versus accuracy trade-off that classical methods are known for.