omega-fold
Use OmegaFold for protein structure prediction from FASTA to PDB, including CLI inference, FASTA/PDB handling, and Python model API inspection.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- omega-fold
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use OmegaFold for protein structure prediction from FASTA to PDB, including CLI inference, FASTA/PDB handling, and Python model API inspection.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/install-and-environment.md3.3 KB
- 参考文档references/repo-provenance.md2.3 KB
- 参考文档references/repo-routing-metadata.json331 B
- 参考文档references/troubleshooting.md3.1 KB
- 脚本scripts/check_omega_fold_environment.py5.9 KB
03 SKILL.md 正文
OmegaFold
Use this repo skill when a task involves OmegaFold, de novo protein structure prediction from amino-acid FASTA input, OmegaFold PDB outputs, the omegafold CLI, or the Python APIs omegafold.make_config, omegafold.OmegaFold, and omegafold.pipeline.
OmegaFold is release inference code. It is not a training, fine-tuning, dataset-building, model-serving, or general PDB-editing toolkit.
Start Here
- For command-line prediction from FASTA to PDB, read inference-cli.
- For FASTA validation, pseudo-MSA tensors, PDB output names, and confidence/B-factor interpretation, read data-and-outputs.
- For programmatic model construction, configs, forward inputs/outputs, and API debugging, read model-api.
- For installation, dependency, Python, Torch, CUDA/MPS, and smoke-check guidance shared across routes, read install and environment.
- For cross-cutting failures, read root troubleshooting, then route to the nearest sub-skill troubleshooting file.
Route by Task
| User task | Read |
|---|---|
Build or validate an omegafold INPUT_FILE.fasta OUTPUT_DIRECTORY command |
sub-skills/inference-cli/SKILL.md |
Decide --model, --device, --weights_file, --subbatch_size, or --num_cycle |
sub-skills/inference-cli/references/inference-workflow.md |
| Check whether the CLI is installed without downloading weights | sub-skills/inference-cli/scripts/omegafold_cli_smoke.py |
| Explain accepted FASTA syntax, ambiguous residues, sequence sorting, or output PDB names | sub-skills/data-and-outputs/references/data-formats.md |
Exercise pipeline.fasta2inputs or pipeline.save_pdb safely |
sub-skills/data-and-outputs/scripts/inspect_fasta_pipeline.py |
Instantiate OmegaFold, call make_config, load weights, or inspect forward |
sub-skills/model-api/references/api-reference.md |
| Inspect API signatures/configs without loading checkpoints | sub-skills/model-api/scripts/inspect_model_api.py |
| Diagnose install/import/backend issues before choosing a route | references/install-and-environment.md and references/troubleshooting.md |
Minimal Install Expectations
OmegaFold distribution metadata is OmegaFold and the import module is omegafold. The package exposes the console script omegafold=omegafold.__main__:main.
For legacy-compatible installations, prefer Python 3.8, 3.9, or 3.10 with Biopython and a Torch build compatible with the target accelerator. The release requirements pin torch==1.12.0+cu113; when using that stack, keep numpy<2 to avoid NumPy ABI warnings or import/runtime failures.
Safe import check:
python - <<'PY'
import omegafold
print(omegafold.make_config(1).struct_embedder)
print(omegafold.make_config(2).struct_embedder)
PY
Safe CLI check:
omegafold --help
Use the bundled scripts/check_omega_fold_environment.py when you need a single no-download environment report before choosing a sub-skill.
Safety Rules
omegafold --help, bundled helper scripts, and API signature inspection are safe no-download checks.- Full
omegafold INPUT_FILE.fasta OUTPUT_DIRECTORYinference may download model weights, allocate a large model, use GPU/MPS/CPU resources, and take substantial time. - Do not run full inference unless the user accepts downloads/runtime or provides a local checkpoint through
--weights_file. - Do not assume FASTA file order equals output order; OmegaFold sorts records by sequence length before processing.
- Do not interpret OmegaFold PDB B-factors as experimental crystallographic B-factors; the CLI writes per-residue confidence multiplied by
100into the B-factor field. - Do not present random no-weight outputs as meaningful predictions; load the matching model 1 or model 2 checkpoint before inference.
Repository Facts
- Primary workflow: de novo structure prediction from one or more FASTA sequences to one PDB per sequence.
- Public CLI:
omegafold INPUT_FILE.fasta OUTPUT_DIRECTORY [options]. - Public Python entry points:
omegafold.make_config,omegafold.OmegaFold,omegafold.pipeline.fasta2inputs,omegafold.pipeline.save_pdb. - Model ids:
1and2only. - Device auto-selection: CUDA, then MPS, then CPU.
- Provenance and staleness baseline: references/repo-provenance.md.
04 同仓库更多 Skills
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.
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.
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.
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.
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.
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).