Efficient Exploration of Chemical Space with Docking and Deep Learning

Abstract

With the advent of make-on-demand commercial libraries, the number of purchasable compounds available for virtual screening and assay has grown explosively in recent years, with several libraries eclipsing one billion compounds. Today’s screening libraries are larger and more diverse, enabling the discovery of more-potent hit compounds and unlocking new areas of chemical space, represented by new core scaffolds. Applying physics-based in silico screening methods in an exhaustive manner, where every molecule in the library must be enumerated and evaluated independently, is increasingly cost-prohibitive. Here, we introduce a protocol for machine learning-enhanced molecular docking based on active learning to dramatically increase throughput over traditional docking. We leverage a novel selection protocol that strikes a balance between two objectives: (1) identifying the best scoring compounds and (2) exploring a large region of chemical space, demonstrating superior performance compared to a purely greedy approach. Together with automated redocking of the top compounds, this method captures almost all the high scoring scaffolds in the library found by exhaustive docking. This protocol is applied to our recent virtual screening campaigns against the D4 and AMPC targets that produced dozens of highly potent, novel inhibitors, and a blind test against the MT1 target. Our protocol recovers more than 80% of the experimentally confirmed hits with a 14-fold reduction in compute cost, and more than 90% of the hit scaffolds in the top 5% of model predictions, preserving the diversity of the experimentally confirmed hit compounds.

Automated discovery of noncovalent inhibitors of SARS-CoV-2 main protease by consensus Deep Docking of 40 billion small molecules

“… 19 Recently, similar approaches to DD that utilize AI to predict docking outcomes have emerged, including but not limited to MolPAL (molecular pool-based active learning), 20 lean-docking, 21 and AutoQSAR/DeepChem models. 22 Importantly, the high-throughput nature of DD can be utilized for docking ultra-large libraries but also enables the simultaneous use of multiple docking programs, facilitating the deployment of stringent consensus protocols, as advocated by the best practices of computer-aided drug design (CADD). 23 In other words, one could safely assume that when multiple independent docking approaches agree on a hit, the result should be associated with experimental activity with higher confidence.…”

Discovery of Kinase and Carbonic Anhydrase Dual Inhibitors by Machine Learning Classification and Experiments

“…Docking simulation was applied not for screening but for understanding the binding modes of active molecules. The physics-based docking screen can work synergistically with the informaticsbased ML classification, particularly for screening the ultra-large library comprising more than several billion molecules [15,16]. ML is faster than conventional docking by several hundred-fold and suited for primary crude screening.…”

“…Computational methods have helped to decode off-targets and polypharmacological effects with the data. The similarity ensemble approach (SEA) is a representative example [14][15][16]. Successful SEA applications have been reported for predicting off-targets of FDA-approved drugs and excipients [17][18][19].…”

Systemic Evolutionary Chemical Space Exploration For Drug Discovery

“…Furthermore, the R square of the DL model from the combined set was 0.85, slightly better than other models trained only by a single generation. It is reasonable to believe the model performance is sufficient for prediction [51][52][53]. The model F was then used for the prediction of the 66,687,173 molecules and 2,094 molecules with predicted scores less than -10.5 kcal/mol were subjected for redocking.…”

“…The success of ultra-large compounds virtual screening contributes to the vigor of the make-on-demand library [2,3]. Furthermore, people train machine learning models to accelerate the speed of virtual screening to balance the tradeoff between accuracy and speed [51]. However, it is still a very tough task to do virtual screening of ultra-large libraries directly on the present hardware.…”