Polaris Challenge: Data-Driven Priors to Improve Docking for Pose Prediction.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate prediction of ligand binding poses is crucial for structure-based drug design.
• Existing methods face challenges in predicting poses reliably, especially for viral targets.

Purpose of the Study:

• To develop and evaluate an open-source computational workflow for predicting ligand binding poses.
• To leverage crystallographic data and fragment information to enhance pose prediction accuracy for antiviral drug design.

Main Methods:

• Ensemble docking using Vina-GPU, augmented with fragment-derived priors for pose generation and scoring.
• Fallback to MM/GBSA for pose scoring when fragment information is unavailable.
• Utilized time-split test sets for evaluating pose prediction performance.

Main Results:

• Achieved over 50% success in predicting ligand poses within 2 Å RMSD for SARS-CoV-2 and MERS-CoV protease targets.
• Fragment-informed rescoring and pose generation significantly improved prediction accuracy.
• Demonstrated successful transfer of knowledge from SARS-CoV-2 to MERS-CoV, highlighting conserved binding pockets.

Conclusions:

• Docking-based workflows augmented with transferable fragment knowledge and physics-based refinements achieve competitive accuracy for pose prediction.
• Conserved binding pockets across related viral proteases facilitate knowledge transfer for pose prediction.
• The developed workflow is efficient, fully open-source, and applicable to protein families.