A Database for Large-Scale Docking and Experimental Results

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• The availability of large compound libraries has advanced molecular docking, but shared results are limited.
• This data scarcity hinders benchmarking of machine learning and chemical space exploration methods.

Purpose of the Study:

• To create a publicly accessible database of large-scale molecular docking campaigns.
• To provide poses, scores, and in vitro results for 6.3 billion docked molecules against 11 targets.
• To facilitate benchmarking and exploration of vast chemical spaces.

Main Methods:

• Developed a website to host data from recent large library docking campaigns.
• Docked 6.3 billion molecules against 11 biological targets.
• Experimentally tested 3,729 selected compounds.
• Trained machine learning models to predict docking scores and identify top-scoring molecules.

Main Results:

• The database provides unprecedented access to docking poses, scores, and experimental validation data.
• Proof-of-concept studies demonstrated that machine learning models improve with larger training datasets.
• Models correlating well with docking scores did not always enrich top-ranked molecules or novel ligands.

Conclusions:

• The new database is a valuable resource for advancing molecular docking and machine learning applications in drug discovery.
• Sharing comprehensive docking data is essential for robust method development and validation.
• Further research with more sophisticated methods is needed to fully leverage this data for ligand discovery.