Stability-Driven Pose Prediction and Ligand Design via Contact Persistence Analysis from Molecular Dynamics...

Molecular dynamics simulations provide deep insights into protein-ligand interactions, but their high computational cost often limits their application in early-stage drug design. To address this issue, we developed a computational workflow that integrates classical docking, high-temperature molecular dynamics simulations at 400 K, and residue-level contact persistence analysis via the R-value metric for a rapid assessment of stable binding poses. By retrospective evaluation using 137 protein-ligand complexes across 10 therapeutically relevant targets with a total of 1.64 ms of aggregate simulation time, we found that 83.9% of top-ranked poses identified by this pipeline exhibited a root-mean-square deviation ≤3.0 Å from experimental structures, indicating structural agreement with experimentally observed binding modes under the evaluation criteria used in this study. Furthermore, the R-value, a normalized measure of intermolecular noncovalent contact persistence, showed a correlation with experimental binding affinities (pK _d), ranging from 0.40 to 0.94 across 31 systems involving 4 targets, suggesting an association between interaction persistence and experimental binding affinity within the analyzed dataset. We examined whether functional group-level contact persistence can identify ligand moieties associated with stable interactions, thereby supporting hypothesis-driven ligand modification strategies based on persistent contacts. To illustrate the potential application of the workflow, we generated in silico derivatives of TH-Z835 and MRTX1133 as KRAS^G12D inhibitors and evaluated them using the R-value framework as a case study, where several derivatives showed higher predicted interaction persistence than the parent compounds. These analyses illustrate how the workflow may assist in prioritizing ligand modifications for further investigation in structure-based drug design.