Electrostatic-field and surface-shape similarity for virtual screening and pose prediction.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Molecular Modeling

Background:

• Accurate 3D molecular similarity is crucial for virtual screening and drug design.
• Existing methods often rely on heuristics or user-defined features, limiting their physical meaningfulness.
• Conformation-dependent molecular electrostatics require robust comparison methods.

Purpose of the Study:

• Introduce eSim, a new method for rapid 3D molecular similarity computation.
• Evaluate eSim's performance in virtual screening and ligand pose prediction.
• Demonstrate eSim's utility for both large-scale screening and rational drug design.

Main Methods:

• eSim combines electrostatic field comparison with molecular surface-shape and directional hydrogen-bonding preferences.
• Utilizes conformation-dependent molecular electrostatics without heuristic 'colors' or feature types.
• Assessed performance on the DUD-E benchmark dataset (102 targets) and curated ~400,000 bound ligand pose pairs.

Main Results:

• eSim achieved a mean ROC area of 0.74 in moderately fast screening mode, outperforming alternate methods.
• Screening performance exceeded a mean ROC area of 0.82 for 92 targets using multiple ligands as similarity targets.
• Achieved an 80% success rate for RMSD < 2 Å in pose prediction, with 70% success for dissimilar scaffolds.

Conclusions:

• eSim provides a rapid and accurate method for 3D molecular similarity computation.
• The approach enables efficient virtual screening and precise ligand pose prediction.
• eSim's physically meaningful comparisons support both large-scale screening and rational drug design.