# Non-InterPro HMM metadata

`non_interpro_hmm_metadata.tsv` contains one normalized row per HMM model for:

- ABCF
- AMR
- AntiDefenseFinder
- DefenseFinder
- NetFlax
- PAT
- PHROG
- SAS
- VFDB

Pfam is intentionally excluded because the pipeline uses the richer InterProScan
member-database metadata for Pfam.

The table was built from the HMM archives in `Desktop/Annotation_pipeline/hmm_db`,
the NetFlax supplementary workbook `netflax_SI.xlsx`, `PAT_HMMs_384_desc.tsv`, and
the PHROG v4 annotation table:

```text
https://phrogs.lmge.uca.fr/downloads_from_website/phrog_annot_v4.tsv
```

AMR and VFDB descriptions come from embedded HMM `DESC` fields. ABCF,
AntiDefenseFinder, DefenseFinder, and SAS use stable model names as their
descriptions. PAT and PHROG use their external annotation tables.

NetFlax roles are derived from antitoxin/toxin positions in the complete predicted
TA-pair sheet. Four nodes (`D127`, `D3`, `M183`, and `M37`) occur in both positions
and are labelled `predicted_toxin_or_antitoxin`. Curated fold-cluster descriptions
are available for 84 of 310 NetFlax nodes.

Rebuild with `annotation-build-hmm-metadata`; the command validates model-name
uniqueness and metadata coverage before writing output.