PMO-Dock: Benchmarking Docking, Specificity, and Generalization in Molecular Optimization
Abstract
The Practical Molecular Optimization (PMO) benchmark standardized evaluation in molecular optimization, but it is built on simple property-based oracles that do not capture structure-based drug design. As the field has moved to docking-based objectives, evaluation practice has become unstandardized, and different methods report results on different docking tasks, often without a shared protocol for fair comparison or sample-efficiency constraints. To address this, we introduce Practical Molecular Optimization for Docking (PMO-Dock), a benchmark and protocol for docking-based optimization consisting of 25 tasks covering hit generation, lead optimization, and a new specificity task requiring strong on-target binding while penalizing off-target interactions. The protocol separates model development from final evaluation: validation tasks are used for hyperparameter tuning, and test tasks are held out for final evaluation and are not used during development. Evaluation on the test tasks is done under a strict oracle budget to reflect realistic drug-discovery settings. We benchmark four diverse high-performing methods spanning different optimization paradigms, Saturn (reinforcement learning), GenMol (discrete diffusion), Genetic-guided GFlowNet, and ChemLlama (LLM-based evolutionary search). Our results show no universally best method across tasks, additional analyses of hyperparameter transfer (cross-task and cross-category) and sensitivity show that transfer is often limited and varies by method, motivating method-dependent selection strategies. This benchmark provides a concrete standard for evaluating generalization to unseen tasks and sample efficiency under a limited test-time oracle budget.
Related
- ELiAH2: update of the atlas of E3 ligases in human tissues by integrating tissue-specificity and docking properties of ligands for targeted protein degradation
- Leakage-controlled benchmarking reveals generalization limits of deep learning for protein-ligand binding affinity prediction
- Stoichiometry-Dependent Specificity in Biotin Enrichment: A Benchmarking Framework for Proximity-Labeling Proteomics.
- Insight into the mechanism of molecular interactions of the NDM metallo-beta-lactamase and cyclic boronate inhibition by molecular docking and molecular dynamics simulations
- Improving Molecular Docking Pose Recovery through Targeted Side-Chain Monte Carlo Sampling of an Apo Structure
BibTeX
@misc{simonyan2026pmodock,
title = {{PMO-Dock: Benchmarking Docking, Specificity, and Generalization in Molecular Optimization}},
author = {Gor Simonyan and Tatevik Abrahamyan and Narek Abelyan and Tigran Fahradyan and Hrant Khachatrian},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.19011},
eprint = {2610.19011},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.19011},
}
RIS
TY - UNPB
TI - PMO-Dock: Benchmarking Docking, Specificity, and Generalization in Molecular Optimization
AU - Gor Simonyan
AU - Tatevik Abrahamyan
AU - Narek Abelyan
AU - Tigran Fahradyan
AU - Hrant Khachatrian
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2610.19011
UR - https://arxiv.org/abs/2610.19011
AB - The Practical Molecular Optimization (PMO) benchmark standardized evaluation in molecular optimization, but it is built on simple property-based oracles that do not capture structure-based drug design. As the field has moved to docking-based objectives, evaluation practice has become unstandardized, and different methods report results on different docking tasks, often without a shared protocol for fair comparison or sample-efficiency constraints. To address this, we introduce Practical Molecular Optimization for Docking (PMO-Dock), a benchmark and protocol for docking-based optimization consisting of 25 tasks covering hit generation, lead optimization, and a new specificity task requiring strong on-target binding while penalizing off-target interactions. The protocol separates model development from final evaluation: validation tasks are used for hyperparameter tuning, and test tasks are held out for final evaluation and are not used during development. Evaluation on the test tasks is done under a strict oracle budget to reflect realistic drug-discovery settings. We benchmark four diverse high-performing methods spanning different optimization paradigms, Saturn (reinforcement learning), GenMol (discrete diffusion), Genetic-guided GFlowNet, and ChemLlama (LLM-based evolutionary search). Our results show no universally best method across tasks, additional analyses of hyperparameter transfer (cross-task and cross-category) and sensitivity show that transfer is often limited and varies by method, motivating method-dependent selection strategies. This benchmark provides a concrete standard for evaluating generalization to unseen tasks and sample efficiency under a limited test-time oracle budget.
ER -