GNINA sampling benchmark outputs for the OpenBind EV-A71 2A protease dataset

This record contains the complete GNINA docking outputs used for the sampling analysis accompanying the Blopig post The cost of a better pose: balancing GNINA sampling and runtime.

The analysis uses the OpenBind EV-A71 2A protease dataset to investigate how pose-recovery performance changes with increased GNINA exhaustiveness, repeated docking using different random seeds, and docking from multiple ligand starting conformers.

The archive contains the original complex-level inputs required for downstream analysis, the prepared docking inputs, and the complete GNINA outputs used in the study. It is intended to allow the pose analysis, runtime analysis, and figures to be reproduced without rerunning the docking calculations.

The corresponding analysis and plotting code is available on GitHub: A71EV2A-blopig.

The original OpenBind EV-A71 2A protease dataset is described in a preprint and is also available separately on Zenodo.

Users wishing to reproduce the complete workflow from the original data, including receptor and ligand preparation and GNINA docking, should start from that release. Users wishing to reproduce only the downstream analysis can use the precomputed docking outputs provided here.

Archive structure

A71EV2A_data/
├── annotated_complexes.csv
└── data/
    ├── <complex_id>/
    │   ├── original complex inputs
    │   ├── prepared_inputs/
    │   └── gnina/
    └── ...

Sampling configurations

The archive contains the outputs corresponding to the sampling configurations used in the accompanying analysis:

• One ligand starting conformer with five independent random seeds
• Five ligand starting conformers using a single random seed
• GNINA exhaustiveness values of 8, 16, and 32

For the multi-run comparisons, eight poses are retained from each seed or starting conformer, giving a 40-pose candidate pool that is globally reranked using CNNscore.

Reproducing the analysis

The GitHub repository contains the scripts required to reanalyse these docking outputs, collect runtime statistics, and regenerate the final figures. It also includes the processed pose-level Parquet files and runtime summaries needed to reproduce the figures directly without downloading this archive.

Repository: https://github.com/Jnelen/A71EV2A-blopig