On the Difficulty to Rescore Hits from Ultralarge Docking Screens

Area of Science:

• Computational chemistry
• Drug discovery
• cheminformatics

Background:

• Ultralarge virtual screening (ULVS) offers higher hit rates for potent ligands compared to conventional methods.
• Efficient postprocessing is crucial for selecting diverse compounds from millions of virtual hits for synthesis and evaluation.

Purpose of the Study:

• To retrospectively analyze the performance of various rescoring methods for binding affinity prediction in ULVS.
• To evaluate the impact of energy refinement on different rescoring approaches.

Main Methods:

• Analysis of ten ULVS hit lists against in vitro binding assays.
• Testing eight rescoring methods: empirical, machine learning, molecular mechanics, and quantum mechanics.
• Assessing the effect of energy refinement prior to rescoring.

Main Results:

• No single rescoring method robustly distinguished active compounds from inactive ones across all assays.
• Sophisticated methods generally performed better, but with limitations.
• Energy refinement improved molecular and quantum mechanics but worsened empirical and machine learning predictions.

Conclusions:

• Current rescoring methods are insufficient for reliably prioritizing compounds from ULVS.
• Further development of robust binding affinity prediction tools is necessary for effective drug discovery pipelines.
• Energy refinement strategies need careful consideration based on the chosen rescoring approach.