OpenBind EV-A71 2A RBFE Benchmark

This repository contains benchmark structures and results from the recent OpenBind EV-A71 2A benchmark. Full data are available from OpenBind; these structures have been selected to benchmark RBFE workflows.

Release Structure

This repository contains two benchmark release folders:

• openbind_ev71_2a_pyrrolidine_benchmark_release/: focused 32-compound pyrrolidine-thiopyrimidine subset.
• openbind_ev71_2a_methylthio_pyrimidine_benchmark_release/: full 76-compound methylthio-pyrimidine benchmark set.

Each folder has the same structure.

receptor/

Contains the prepared EV-A71 2A protease receptor used for docking/FEP setup.

• ev71_2a_x7339a_template_prepared.pdb: prepared receptor template derived from the A71EV2A-x7339a holo structure.

ligands/

Contains ligand pose files in SDF format.

• *_rowan_docked_poses.sdf: Rowan analogue-docked top-ranked poses, one pose per compound.
• *_xtal_poses.sdf: representative crystallographic poses, one pose per compound. Use this file to run FEP from crystal poses.
• *_xtal_all_crystal_instances.sdf: all available crystallographic ligand instances for the selected compounds, including duplicate crystal structures where present.

subset/

Contains compound identity and affinity metadata.

• *_subset.csv: compound IDs, SMILES, experimental affinities, source crystal structures, and source ligand reference files.
• *_subset.smi: SMILES and compound IDs, suitable for docking or cheminformatics workflows.

results/

Contains Rowan FEP results (to date). These results are provided as a baseline and are not yet "good."

• rowan_results_per_compound_long.csv: one row per compound per Rowan run.
• rowan_results_per_compound_wide.csv: one row per compound, with prediction columns for each Rowan run.
• rowan_results_per_edge_long.csv: one row per graph edge per Rowan run.
• rowan_results_per_edge_wide.csv: one row per markdown

Release Folder Structure

Each release folder contains a self-contained benchmark subset with receptor coordinates, ligand poses, compound metadata, and Rowan RBFE outputs.

receptor/

Prepared protein structure used as the receptor template for the benchmark.

• ev71_2a_x7339a_template_prepared.pdb: prepared EV-A71 2A protease receptor, derived from the A71EV2A-x7339a holo structure.

ligands/

Ligand pose files in SDF format.

• *_rowan_docked_poses.sdf: Rowan analogue-docked top-ranked pose for each compound.
• *_xtal_poses.sdf: one representative crystallographic ligand pose per compound. Use this file for FEP graphs starting from crystal poses.
• *_xtal_all_crystal_instances.sdf: all available crystallographic ligand instances for the selected compounds, including duplicate crystal structures where present.

subset/

Compound-level metadata for the release subset.

• *_subset.csv: compound IDs, SMILES, experimental affinity values, source crystal structure paths, and duplicate/crystal-instance mappings.
• *_subset.smi: SMILES and compound ID, suitable for docking or cheminformatics workflows.

results/

Processed Rowan RBFE outputs.

• rowan_results_per_compound_long.csv: one row per compound per Rowan run.
• rowan_results_per_compound_wide.csv: one row per compound, with prediction columns for each Rowan run.
• rowan_results_per_edge_long.csv: one row per RBFE edge per Rowan run.
• rowan_results_per_edge_wide.csv: one row per RBFE edge, with columns for each Rowan run.
• rowan_results_overall_by_run.csv: aggregate performance metrics by run, including RMSE, MAE, rank correlations, and linear-fit statistics.

License

Code in this repository is licensed under the MIT license.

The original OpenBind crystallographic and binding-affinity data were obtained from Zenodo and are licensed under the CC0 license.

Corin Wagen, with assistance from GPT 5.5