Retrieval Augmented Docking Using Hierarchical Navigable Small Worlds.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Make-on-demand chemical libraries, such as the ZINC-22 library with nearly 6.4 billion molecules, have expanded molecular docking capabilities.
• The immense size of these libraries presents significant computational challenges, making exhaustive docking infeasible for most researchers.

Purpose of the Study:

• To introduce and validate a novel method, retrieval augmented docking (RAD), for efficiently searching large chemical spaces.
• To demonstrate RAD's ability to recover a high percentage of virtual active compounds while significantly reducing the number of molecules that need to be docked.

Main Methods:

• Organizing and traversing chemical space using hierarchical navigable small-world graphs.
• Applying the retrieval augmented docking (RAD) method to large-scale docking campaigns against specific protein targets (D4 and AmpC) and a diverse set of proteins (DUDE-Z).

Main Results:

• RAD successfully recovered 95% of virtual actives for D4 and AmpC targets after docking only 10% of the respective libraries (99.5 million and 138 million molecules).
• Across 43 DUDE-Z proteins, RAD recovered an average of 87% of virtual actives while docking only 10% of the evaluated molecules (50.3 million associations).
• The method demonstrated protein-agnostic applicability and maintained chemical diversity without sacrificing performance.

Conclusions:

• Retrieval augmented docking (RAD) offers a computationally efficient solution for exploring massive chemical libraries in drug discovery.
• RAD significantly reduces the computational burden of virtual screening, making large-scale docking more accessible.
• The protein-agnostic nature of RAD allows for flexible and cost-effective docking campaigns against multiple targets.