A database for large-scale docking and experimental results.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• The past six years have seen a significant increase in accessible compounds, enhancing molecular docking techniques.
• Sharing docking campaign results is crucial for benchmarking machine learning (ML) and chemical space exploration methods.

Purpose of the Study:

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

Main Methods:

• Developed a website (lsd.docking.org) to host docking campaign data.
• Docked 6.3 billion molecules against 11 targets.
• Experimentally tested 3729 compounds.
• Trained ML models to predict docking scores and identify top-scoring molecules.

Main Results:

• Models trained on larger datasets showed improved performance.
• High correlation between ML predictions and docking scores did not always lead to successful enrichment of effective ligands.
• Even top-ranked molecules from docking did not guarantee experimental success, highlighting limitations in current predictive models.

Conclusions:

• The new database supports ML model training and chemical space exploration.
• Predictive models need further refinement as high docking scores do not always translate to experimental validation.
• Openly sharing large-scale docking data is essential for advancing drug discovery methodologies.