Assessing small molecule conformational sampling methods in molecular docking.

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Conformational sampling is crucial for accurate molecular docking.
• Various algorithms exist, but their impact on docking performance is not fully understood.

Purpose of the Study:

• To evaluate the performance of six traditional and one deep learning-based small molecule conformational sampling method in molecular docking.
• To assess binding pose reproducibility and screening power using diverse benchmark datasets.

Main Methods:

• UCSF DOCK 3.7 software was used for molecular docking.
• Six traditional methods (Omega, BCL::Conf, CCDC Conformer Generator, ConfGenX, Conformator, RDKit ETKDGv3) and Torsional Diffusion (deep learning) were employed.
• Docking was performed against Platinum Diverse, PoseBusters, and DUDE-Z datasets.

Main Results:

• Different conformational sampling methods demonstrated varying docking performance.
• Performance differences are attributed to method-specific sampling preferences, like dihedral angle ranges.
• No single method universally outperformed others across all metrics.

Conclusions:

• The choice of conformational sampling method significantly impacts molecular docking outcomes.
• Combining complementary sampling strategies may offer improved docking accuracy and screening power.
• Further research into ensemble-based sampling is warranted for drug discovery applications.