Answers are grounded in this paper and cite their sources.
Benchmark dataset captures real-world drug discovery challenges and supports FEP evaluation in chemical systems.
Free Energy Perturbation (FEP) is emerging as one of the most accurate and powerful methods for predicting binding affinities in structure-based drug discovery. While several benchmark datasets have been developed to assess FEP performance, including one with 1867 ligands, these datasets are often simplified and curated to match current methodological capabilities. As a result, they do not reflect the chemical diversity or complexity typically encountered in real-world drug discovery projects, limiting their ability to reveal the full potential and practical limitations of FEP. To address this issue, we introduce Uni-FEP Benchmarks, a large-scale and publicly available dataset constructed from drug discovery cases curated from the ChEMBL database. Through careful selection and processing, we identified approximately 1000 protein–ligand systems that are suitable for FEP calculations. These systems include around 40,000 ligands, making this dataset significantly larger than any previously available FEP benchmark. It captures a wide range of chemical challenges such as scaffold replacements, charge changes, and other modifications representative of real medicinal chemistry efforts. We evaluated the benchmark using Uni-FEP workflow, an automated and scalable FEP workflow developed at ATOMBEAT. This platform enables consistent and reproducible FEP simulations starting from minimal inputs. Uni-FEP Benchmarks is fully open to the scientific community (https://github.com/dptech-corp/Uni-FEP-Benchmarks), providing a valuable resource for both methodological development and applied drug discovery research. It supports transparent evaluation of FEP workflows under realistic conditions and serves as a foundation for improving accuracy, robustness, and applicability of FEP in practical settings.
No takes yet. Share an insight, caveat, or question.
Zou et al. (2025) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: