Sofiene Khiari | Searcharxiv

arXiv2026

AgenticPosesRanker: An Agentic AI Framework for Physically Grounded Ranking of Protein-Ligand Docking Poses

Scoring functions remain the principal bottleneck in molecular docking: they routinely fail to rank near-native poses above decoys, and their composite single-score design obscures the physicochemical basis of each ranking error. We present AgenticPosesRanker, an agentic AI framework that combines six deterministic, physically grounded analysis tools (interaction fingerprinting, solvent-accessible burial, conformational strain, steric-clash detection, unsatisfied-polar-atom penalty, and chemical-identity extraction) with large-language-model (GPT-5) chain-of-thought reasoning to evaluate and rank docking poses. On a curated benchmark of ten protein-ligand systems (162 poses) balanced by construction between Smina scoring-function successes and failures, the agent achieved 50.0% best-pose accuracy, matching the design-fixed Smina baseline of 50.0% and significantly exceeding a 7.7% uniformly random baseline (p < 0.001, one-sided exact binomial test). The balanced-benchmark accuracy decomposes symmetrically: the agent retained 80% (4/5) of the Smina-success systems and recovered 20% (1/5) of the Smina-failure systems, so the aggregate 50% reflects one regression offset by one recovery rather than any net improvement over the Smina reference. Decision-attribution analysis showed high alignment between the agent's self-reported tool weights and objective metric separations of the selected pose (median \r{ho} = +0.83), consistent across correct and incorrect outcomes, localising the performance ceiling to tool-suite coverage rather than reasoning inconsistency. These results establish a methodological template for evaluating agentic AI against objective ground truth in the natural sciences and position the framework as an interpretable curation layer for late-stage pose refinement in structure-based drug design.

q-bio.BM↗

arXiv2025

Synthetic Protein-Ligand Complex Generation for Deep Molecular Docking

The scarcity of experimental protein-ligand complexes poses a significant challenge for training robust deep learning models for molecular docking. Given the prohibitive cost and time constraints associated with experimental structure determination, scalable generation of realistic protein-ligand complexes is needed to expand available datasets for model development. In this study, we introduce a novel workflow for the procedural generation and validation of synthetic protein-ligand complexes, combining a diverse ensemble of generation techniques and rigorous quality control. We assessed the utility of these synthetic datasets by retraining established docking models, Smina and Gnina, and evaluating their performance on standard benchmarks including the PDBBind core set and the PoseBusters dataset. Our results demonstrate that models trained on synthetic data achieve performance comparable to models trained on experimental data, indicating that current synthetic complexes can effectively capture many salient features of protein-ligand interactions. However, we did not observe significant improvements in docking or scoring accuracy over conventional methods or experimental data augmentation. These findings highlight the promise as well as the current limitations of synthetic data for deep learning-based molecular docking and underscore the need for further refinement in generation methodologies and evaluation strategies to fully exploit the potential of synthetic datasets for this application.

q-bio.BM↗