evaluation-benchmarks
Plan and troubleshoot DiffDock benchmark evaluation runs, RMSD/confidence metrics, GNINA post-processing, and vendored spyrmsd checks.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- evaluation-benchmarks
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Plan and troubleshoot DiffDock benchmark evaluation runs, RMSD/confidence metrics, GNINA post-processing, and vendored spyrmsd checks.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/evaluation-workflows.md9.2 KB
- 参考文档references/rmsd-and-metrics.md7.9 KB
- 参考文档references/troubleshooting.md9.2 KB
03 SKILL.md 正文
DiffDock Evaluation Benchmarks
Use this sub-skill when a task is about reproducing or diagnosing DiffDock benchmark evaluations rather than making ad-hoc docking predictions.
Route Here For
- Building benchmark commands for PDBBind, BindingMOAD/DockGen, or PoseBusters with
python -m evaluate. - Checking required dataset, split, ESM embedding, score model, confidence model, sampling, and output flags before an expensive run.
- Explaining saved evaluation outputs such as
rmsds.npy,confidences.npy,centroid_distances.npy,gnina_metrics.pkl, and printed success-rate metrics. - Diagnosing symmetry-corrected RMSD, confidence re-ranking, GNINA minimization, cache, CUDA/CPU, and benchmark data-layout failures.
- Inspecting the vendored
spyrmsdCLI availability without running a benchmark.
Route Elsewhere
- For single-complex or CSV-driven docking prediction, use docking-inference.
- For dataset preparation, ESM embedding generation, training inputs, or training runs, use training-data.
Workflow
- Identify the benchmark family: PDBBind, DockGen/BindingMOAD, or PoseBusters.
- Read evaluation workflows for dataset layouts, required flags, and command patterns.
- Use build_evaluation_command.py to construct a safe
python -m evaluatecommand; the helper prints only and never runs DiffDock. - Read RMSD and metrics before interpreting arrays, confidence ordering, GNINA metrics, or symmetry-corrected RMSD results.
- Use troubleshooting for dependency, backend, dataset, split, ESM, GNINA, cache, OOM, and result interpretation failures.
- Use inspect_spyrmsd_cli.py to check vendored
spyrmsdavailability or construct a reference RMSD CLI command.
Safety Notes
- Benchmark evaluation imports a heavy optional stack including Torch, PyG, W&B, RDKit, model code, and dataset loaders; plan validation before running it.
- Full benchmark runs require prepared data, cached ESM embeddings, model checkpoints, and significant compute; classify native benchmark commands as expensive unless scoped with
--limit_complexes. - GNINA is an external executable invoked by DiffDock; verify the binary and receptor/ligand layout before enabling GNINA flags.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.