The CACHE Challenges are a series of challenges to test the performance of computational methods for finding small molecules that bind to a given protein target. CACHE stands for Critical Assessment of Computational Hit-finding Experiments. You can think of it as an analogue to the CASP challenges but for small molecule drug discovery.
During my PhD I have participated in the first two challenges which targeted the Parkinson’s-associated LRRK2 protein and the SaRs-CoV-2 nsp13 protein. My submissions for both challenges succesfully resulted in the identification of multiple novel, experimentally validated hit and potential hit compounds.
For both challenges, my strategy was largely based on using gnina, an open-source molecular docking program developed by the Koes lab which uses a deep-learning based scoring function. In this first challenge, I docked the entire MolPort database, approximately 7 million molecules into the target receptor. In the second challenge, I wanted to screen the Enamine REAL database, which contains 43 billion molecules. Since docking all 43 billion molecules is infeasible, I instead implemented an active learning strategy to intelligently search through the database for hit compounds. I only had to dock 600,000 molecules, and of the top ~50 that were tested, 5 were experimentally identified as potential hits.