Improving Molecular Docking Pose Recovery through Targeted Side-Chain Monte Carlo Sampling of an Apo Structure

Abstract

Receptor flexibility complicates molecular docking when only an unbound (apo) protein structure is available. Local side-chain rearrangements can alter binding-pocket accessibility and ligand complementarity, motivating approaches that sample these motions efficiently. Here, we evaluate a targeted Markov Chain Monte Carlo (MCMC) protocol that combines local and larger side-chain torsional proposals with periodic molecular dynamics (MD) relaxation under harmonic backbone restraints. Receptor ensembles generated from an apo structure were evaluated against 859 experimental complexes from the OpenBind 2A protease benchmark and compared with the apo receptor and ensembles generated by restrained and unrestrained MD. With regular-space representative selection, MCMC recovered a ligand pose within 2 [A] of the experimental pose for 74.4% of complexes, compared with 51.0% for apo and 59.5% for restrained MD. Median minimum ligand heavy-atom RMSD decreased from 1.96 [A] for apo and 1.82 [A] for restrained MD to 1.62 [A] for MCMC. The largest gains occurred for complexes with greater side-chain displacement relative to pocket backbone displacement, consistent with the motions targeted by the protocol. MCMC representatives retained pocket geometries within the experimentally observed range, whereas unrestrained-MD representatives exhibited reduced pocket free volume and poor recovery. These findings support targeted side-chain sampling as a practical approach to generating apo-derived docking ensembles. The reported improvements measure best-case pose coverage across generated poses and receptors; their extension to prospective pose ranking and virtual screening remains to be established.