Molecular Libraries
Podium
.jpg "John Irwin, PhD photo")
John Irwin, PhD
Adjunct Profesor
UCSF
Despite intense interest in expanding chemical space, libraries of hundreds-of-millions to billions of diverse molecules have remained inaccessible. In principle, structure-based docking can address such large libraries, but to do so the molecules must be readily obtained, the computation must be tractable, and new potential ligands must outscore the inevitable decoys. Here, we investigate docking screens of over 170 million make-on-demand, lead-like compounds. The molecules derive largely from 109 well-characterized two-component reactions with a >85% synthesis success rate. The resulting library is diverse, representing over 10.7M scaffolds not found in “in-stock” commercial collections. In benchmarking well-behaved targets, the enrichment of ligands versus decoys improved with library size, suggesting that more ligand-like molecules exist to be found as the library grows. To test this prospectively, 99 and 138 million molecules were docked against the soluble enzyme AmpC β-lactamase and the membrane-bound D4 dopamine receptor respectively. From among the top-ranking docking hits for AmpC, 44 diverse molecules were synthesized and tested. Five inhibited, including an unprecedented 1.3 uM phenolate that is the most potent AmpC inhibitor found from any screen. Within-library optimization revealed a 110 nM analog, the most potent non-covalent AmpC inhibitor known. Crystal structures of these inhibitors confirmed their fidelity to the docking prediction. For the dopamine receptor, 549 molecules were synthesized and tested from among top docking ranks, and also from intermediate and low ranks. On testing, hit rates fell monotonically with score, ranging from 24% for the highest ranking, declining through intermediate scores, and dropping to a 0% hit rate for the lower ranks. Integrating across the resulting hit-rate curve predicts 481,000 D4 active molecules in 72,600 scaffolds. Of the 81 new D4 actives found here, 30 had Ki values < 1 uM. The most potent was a 180 pM Gi-biased, selective, full agonist, among the most potent sub-type selective agonists known for this receptor. Prospects for making over 1 billion lead-like molecules community-accessible, and for their prioritization in structure-based screens, will be considered.