Artificial intelligence-enabled virtual screening of ultra-large chemical libraries with deep docking

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• The exponential growth of chemical libraries necessitates advanced virtual screening methods.
• Current structure-based virtual screening struggles with billion-molecule scale libraries.

Purpose of the Study:

• To present a generalized Deep Docking (DD) protocol for efficient screening of ultra-large chemical libraries.
• To enable rapid identification of potential drug candidates from massive datasets.

Main Methods:

• The Deep Docking (DD) platform combines structure-based docking with iterative ligand-based predictions.
• The protocol involves library preparation, receptor preparation, sampling, docking, model training, inference, and residual docking.
• An iterative workflow refines predictions and augments training data.

Main Results:

• DD achieves up to 100-fold acceleration in virtual screening.
• Hundreds- to thousands-fold enrichment of virtual hits is observed with minimal loss of candidates.
• The protocol enables screening of billion-molecule libraries using standard computational resources.

Conclusions:

• The generalized DD protocol is a versatile and automatable solution for computer-aided drug discovery.
• This open-source protocol significantly accelerates the exploration of vast chemical spaces.
• DD facilitates efficient drug candidate identification from ultra-large chemical libraries.