AGENT SKILL

alphafold2

Guide alphafold2-pytorch protein sequence, MSA, distogram, angle-logit, coordinate-refinement, recycling, embedding, and structure-utility workflows with version-aware troubleshooting.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
alphafold2
license
MIT
metadata
{ "disco-role": "operating" }
description
Guide alphafold2-pytorch protein sequence, MSA, distogram, angle-logit, coordinate-refinement, recycling, embedding, and structure-utility workflows with version-aware troubleshooting.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

alphafold2-pytorch

Use this skill when a task names alphafold2-pytorch, alphafold2_pytorch, Alphafold2, Evoformer, protein sequence/MSA folding, distograms, angle logits, residue coordinates, ESM/MSA/ProtTrans embeddings, recycling, MDS, or protein-structure metrics.

This graph targets the unofficial PyTorch implementation at distribution version 0.4.32. It is not the official DeepMind AlphaFold2 implementation. Its current source and README disagree in several places; prefer the verified current API described by the linked sub-skills over older README recipes.

Install and inspect

Install the public distribution, then verify the import before building a model:

python -m pip install alphafold2-pytorch==0.4.32
python -c "import alphafold2_pytorch; print(alphafold2_pytorch.Alphafold2)"

Coordinate and utility paths also need the scientific and geometric requirements declared by the distribution, including a mutually compatible PyTorch/PyTorch3D pair, invariant-point-attention, sidechainnet, mdtraj, ProDy, mp-nerf, and OpenMM where the selected helper imports it. Use the environment checker for a read-only summary; use cross-cutting troubleshooting when dependency resolution selects an incompatible backend.

Route by task

  • Core sequence/MSA trunk, Evoformer, distograms, angle logits, masks, templates, and current constructor/forward contracts: read core-model.
  • Residue coordinates, invariant-point refinement, confidence, auxiliary returns, and recycling: read structure-and-recycling.
  • ESM, MSA Transformer, ProtTrans wrappers or safe precomputed representations: read embeddings.
  • Distogram-to-distance conversion, MDS, atom masks, sidechain layouts, Kabsch, LDDT, GDT, TM-score, and distance losses: read utilities.

Start with the owning sub-skill, then follow its API reference and troubleshooting file. Cross-links are deliberate: do not duplicate a sibling's full API table in the root router.

Operating constraints

  • Use tiny synthetic tensors first. The model is quadratic in sequence length for several trunk operations, and untrained outputs are not scientific structure predictions.
  • At this version, normal MSA input has shape (B, M, N) with the same residue width N as seq; masks should be boolean and on the same device.
  • CPU is the default verification backend. CUDA is an optional acceleration path and must be checked on the actual host; a visible CUDA installation is not proof that a shared device has enough memory.
  • Pretrained embedding wrappers may download code or weights and may require caches, network access, Hugging Face/torch.hub support, or fused operations. Do not trigger those side effects without explicit approval.
  • Full training, multi-terabyte MSA acquisition, DeepSpeed sparse attention, PyRosetta relaxation, and notebook-scale experiments are intentionally not part of this runtime graph. See limitations.

Provenance and refresh

Read repository provenance before treating this graph as current for another checkout. Refresh it when the commit, package version, public signatures, or evidence paths change. The generated runtime graph is self-contained and does not require the source checkout to remain available.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351