Machine Learning-Boosted Docking Enables the Efficient Structure-Based Virtual Screening of Giga-Scale Enumerated...

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Ultra-large screening libraries present challenges for traditional docking-based virtual screening.
• Existing methods struggle with the scale of giga-scale compound libraries.

Purpose of the Study:

• To evaluate the HASTEN tool for accelerating virtual screening of giga-scale libraries.
• To assess HASTEN's efficiency in recalling top-scoring compounds for antibacterial and antiviral targets.

Main Methods:

• Generated a brute-force docking baseline for 1.56 billion compounds using the Glide high-throughput virtual screening protocol.
• Applied HASTEN, a machine learning-boosted strategy, to screen a small fraction (1%) of the library.
• Investigated the impact of hydrogen bonding constraints on docking and ML predictions.

Main Results:

• HASTEN achieved 90% recall of the top 1000 virtual hits by docking only 1% of the library for both antibacterial and antiviral targets.
• Reduced required docking experiments by 99%, significantly shortening screening time.
• Identified optimization potential in handling failed docking attempts for ML-boosted screening.

Conclusions:

• HASTEN is a fast and robust tool for screening giga-scale libraries in drug discovery.
• ML-boosted strategies offer significant advantages in throughput and efficiency over brute-force docking.
• HASTEN facilitates unlocking vast chemical space for drug discovery campaigns.