Bento

Bento is a benchmark repository for evaluating protein-ligand docking methods across curated datasets and prediction outputs.

Highlights

• Unified benchmark assets for ML and classical docking methods.
• Legacy analysis pipeline preserved in scripts/.
• UV-first workflow for local development and CI.
• Reproducibility and HPC execution notes.

Repository Layout

bento/
├── annotated_ligands/          # Train/test tables with ligand-level annotations
├── annotations/                # Annotation maps and ligand class assets
├── datasets/                   # Core benchmark datasets
├── predictions_full_raw/       # Raw predictions from docking methods
├── scripts/                    # Legacy pipeline scripts (01..04)
├── similarity_scores/          # Similarity score artifacts
├── src/bento/                  # Bento CLI wrapper and environment tooling
├── tests/                      # Automated tests
├── docs/                       # Reproducibility and HPC docs
└── slurm/                      # Example SLURM job templates

Quick Start (UV)

1. Sync environment

uv sync --no-editable

--no-editable is recommended to guarantee the bento console entrypoint is available in all Python 3.12 environments.

For development checks:

uv sync --no-editable --extra lint --extra test

For ligand annotation dependencies:

uv sync --no-editable --extra annotation

For pocket similarity dependencies (Linux/HPC):

uv sync --no-editable --extra similarity

2. Validate environment

uv run --extra annotation bento check-env --profile annotation
uv run bento check-env --profile similarity --glosa-dir /path/to/glosa_v2.2

3. Run pipeline commands

Ligand annotation:

uv run --extra annotation bento annotate-ligands \
  --dataset-file datasets/tests.tsv \
  --output-dir outputs/

Pocket similarity:

uv run bento compute-pocket-similarity \
  --data-csv test_run/path_tests.tsv \
  --protein-path path_protein \
  --ligand-path path_ligand \
  --bs-dir bs \
  --glosa-dir /path/to/glosa_v2.2 \
  --output-file similarity_scores/test_pocket_scores.tsv

Annotation mapping:

uv run bento map-annotations \
  --tests-file datasets/tests.tsv \
  --annotations-dir annotations \
  --output-tests-file datasets/tests_annotated.tsv \
  --output-tests-exploded-file datasets/tests_exploded_annotated.tsv

Configuration

Legacy scripts use environment variables:

• BENTO_WORKDIR: repository root path (default: current repository root).
• BENTO_DATABASES_DIR: root path for external dataset files (default: BENTO_WORKDIR).
• BENTO_GLOSA_DIR: path to GLoSA directory (default: <repo>/external/glosa).
• BENTO_REPO_ROOT: optional override for CLI location of legacy scripts.

Example:

export BENTO_WORKDIR=/path/to/Bento
export BENTO_DATABASES_DIR=/path/to/datasets
export BENTO_GLOSA_DIR=/path/to/glosa_v2.2

External Tool Requirements

Some steps require non-Python tools:

• PyMOL (for pocket extraction in script 02; installed with --extra similarity on Linux/HPC).
• Java JDK and g++ (for GLoSA tooling).
• GLoSA v2.2 executable and AssignChemicalFeatures class.

Build GLoSA once inside your BENTO_GLOSA_DIR:

g++ -c glosa.cpp
g++ -o glosa glosa.o
javac AssignChemicalFeatures.java

These tools are validated by:

uv run bento check-env --profile similarity --glosa-dir /path/to/glosa_v2.2

Quality Checks

uv run ruff check .
uv run ruff format --check .
uv run pytest

Reproducibility and HPC

• Reproducibility guide: docs/reproducibility.md
• HPC setup: docs/hpc.md
• SLURM templates: slurm/

Citation

@software{bento_benchmark,
  title={Bento: A Comprehensive Benchmark for Protein-Ligand Docking Methods},
  author={LigandPro Team},
  year={2024},
  url={https://github.com/LigandPro/Bento}
}

License

MIT. See LICENSE.