Add the TM-Vec 2s CATH S95 v4.4.0 index (v1.0)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

README.md

@@ -1,3 +1,74 @@
 ---
-license: apache-2.0
+license: cc-by-4.0
+pretty_name: S(H)ARP reference data
+tags:
+- biology
+- protein
+- cath
+- tm-vec
+- igem
 ---
+
+# S(H)ARP reference data
+
+Reference data for [S(H)ARP](https://gitlab.igem.org/2026/software/usp-brazil/sharp)
+(*Streptomyces* Hidden Antibiotic Regulated Pathways), iGEM USP-Brazil 2026.
+The S(H)ARP repository does not ship these files; its `scripts/fetch_data.sh`
+and its Colab notebook download them.
+
+| File | Contents |
+|---|---|
+| `sharp-tmvec-cath-v1.0.tar.gz` | TM-Vec 2s index of the CATH S95 v4.4.0 domains, unpacks to `sharp/data/tmvec_indexes/cath_s95.*` |
+| `sharp-tmvec-cath-v1.0.tar.gz.sha256` | its SHA-256 checksum |
+
+## Versions
+
+Each change to this repository is a commit. Use a URL pinned to a commit, not
+to `main`: it always serves the same files, and the commit's date shows when
+they were uploaded.
+
+```
+https://huggingface.co/datasets/EdwardNSeven/sharp-data/resolve/<commit>
+```
+
+## Use
+
+In the S(H)ARP Colab notebook, set `TMVEC_INDEX_SOURCE` in *Advanced
+parameters* to the URL above. With a local install:
+
+```bash
+SHARP_DATA_SOURCE=https://huggingface.co/datasets/EdwardNSeven/sharp-data/resolve/<commit> \
+    scripts/fetch_data.sh --tmvec-only
+```
+
+Both check the archive against its `.sha256` before unpacking it.
+
+## How the index was made
+
+- Sequences: `cath-domain-seqs-S95-v4_4_0.fa` from
+  <https://download.cathdb.info/cath/releases/all-releases/v4_4_0/sequence-data/>,
+  73,535 domains (SHA-256 `f93265164dc65340feebfbbffb711fe32bf326037c2bb8422f9fa432e2637387`).
+- Model: the TM-Vec 2s student model, `tmvec2_student.pt`
+  (SHA-256 `fa79c537849f882a805cd583dd1022ed6c927aae55ab8ba80e89b6ad6f1ea76f`), from
+  [tmvec-bench](https://github.com/paarth-b/tmvec-bench).
+- Command: `tmvec2s-search build` from S(H)ARP's `sharp/tmvec2s_search.py`,
+  with sequences over 600 residues split into overlapping windows and pooled
+  (`--long pool --overlap 200`). One 512-dimension float32 vector per domain.
+
+The S(H)ARP Colab notebook can also rebuild the index from the same sequences
+(`TMVEC_INDEX_SOURCE = build`).
+
+## License and citation
+
+The index is derived from CATH, which is licensed under the
+[Creative Commons Attribution 4.0 International License](https://creativecommons.org/licenses/by/4.0/).
+
+- CATH: Waman, V. P., et al. (2025). CATH v4.4: major expansion of CATH by
+  experimental and predicted structural data. *Nucleic Acids Research*,
+  53(D1), D348–D355. [doi:10.1093/nar/gkae1087](https://doi.org/10.1093/nar/gkae1087)
+- TM-Vec 2s: Keluskar, A., Batra, P., Bezshapkin, V., Morton, J. T., &
+  Zhu, Q. (2026). TM-Vec 2s: Accelerated Protein Remote Homology
+  Detection. *bioRxiv*. [doi:10.64898/2026.02.05.704073](https://doi.org/10.64898/2026.02.05.704073)
+- TM-Vec: Hamamsy, T., et al. (2024). Protein remote homology detection
+  and structural alignment using deep learning. *Nature Biotechnology*,
+  42(6), 975–985. [doi:10.1038/s41587-023-01917-2](https://doi.org/10.1038/s41587-023-01917-2)

sharp-tmvec-cath-v1.0.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8b171dbde25ddff3f4e10a294a4ced87949e97faa1a85cdf8c858a7cc51e659
+size 138391879

sharp-tmvec-cath-v1.0.tar.gz.sha256

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0fa11b3da54731b674f106003880151cd9e9ddeca6af06aaf805820645b0275
+size 95