| --- |
| license: cc-by-4.0 |
| tags: |
| - protein-nmr |
| - chemical-shifts |
| - torsion-angles |
| --- |
| |
| # FACET model weights |
|
|
| Model parameters and reference data for [FACET](https://github.com/maxzinke/facet-nmr), |
| which predicts protein backbone phi/psi torsion angles from NMR chemical shifts. |
|
|
| These files are **downloaded automatically on first use** — you do not need to fetch |
| them by hand. `facet/assets.py` resolves them into `~/.facet/` and verifies each |
| against a pinned SHA-256. |
|
|
| They live here rather than in the Python wheel because `facet_retrieval_index.npz` is |
| 133 MB, past PyPI's 100 MB per-file limit; a bundled wheel could not be uploaded at |
| all. Hosting them separately also means a corrected BMRB entry can reach users without |
| a new package release. |
|
|
| ## Contents |
|
|
| | File | Size | What it is | |
| |---|---|---| |
| | `facet_v3.pt` | 5.2 MB | Encoder weights (PyTorch) | |
| | `facet_v3.onnx` | 5.2 MB | The same encoder, ONNX | |
| | `facet_retrieval_index.npz` | 132.6 MB | 254K residue embeddings with phi/psi and labels | |
| | `facet_retrieval_index.entries.json` | 2.0 MB | Per-row source identifiers | |
| | `facet_shift_reference.npz` | 9.6 MB | Mask-safe retrieval reference (optional; absence degrades to the parametric head) | |
|
|
| ## Provenance and licence |
|
|
| These files are licensed **CC BY 4.0**: use them for anything, including commercially, |
| provided you credit the project (see `CITATION.cff` in the source repository). |
|
|
| Structural data (phi/psi, secondary structure) come from the **Protein Data Bank** and |
| chemical-shift data from the **BMRB** — both released under **CC0 1.0**, a |
| public-domain dedication with no conditions. Trained parameters are the work of this |
| project. See `DATA_PROVENANCE.md` in the source repository. |
|
|
| No deposition from the 745-entry benchmark test set is present in the retrieval index |
| or the shift reference (`benchmarks/check_leakage.py` in the source repository). |
|
|
| ## Citation |
|
|
| Zinke, M. ([ORCID 0000-0002-0541-5139](https://orcid.org/0000-0002-0541-5139)). |
| *FACET: backbone torsion angle prediction from NMR chemical shifts* (2026). |
| Software: <https://github.com/maxzinke/facet-nmr> — DOI [10.5281/zenodo.22190034](https://doi.org/10.5281/zenodo.22190034). |
|
|
| Citing BMRB is appreciated: |
| Hoch *et al.*, *Nucleic Acids Research* **51**, D368 (2023), doi:10.1093/nar/gkac1050. |
|
|
