# ECOD classification of the *Salmonella enterica* pangenome (AlphaFold models)

ECOD domain classifications for representative proteins of the *Salmonella
enterica* pangenome, with flags for proteins over-represented in clinical
isolates ("potentially pathogenic"). This is the data set behind:

> Medvedev KE, Zhang J, Schaeffer RD, Kinch LN, Cong Q, Grishin NV (2024).
> Structure classification of the proteins from *Salmonella enterica*
> pangenome revealed novel potential pathogenicity islands. *Sci Rep* 14:12260.
> https://doi.org/10.1038/s41598-024-60991-x

It used to be served as a website (`prodata.swmed.edu/ecod/index_salm.php`).
That site has been retired, and this record is now the archival copy of the
classification and of the database behind it.

## How it was made

Proteins from 79,758 *S. enterica* genomes (NCBI) were clustered with MMseqs2
(50% identity, 80% coverage). Clusters present in at least 80 proteomes were
kept. Each cluster's representative protein got a structure from one of two
sources:
- an AlphaFold DB model, where one matched (at least 90% identity and 85% coverage)
- otherwise, a new AlphaFold prediction

The structures were partitioned and classified with DPAM against ECOD.
Clusters enriched in clinical isolates relative to food and environmental ones
(Fisher's exact test, Holm–Bonferroni p < 0.05) were called potentially
pathogenic. See the paper's Materials and Methods.

## Counts

| | |
|---|---|
| Domains | 17,234, all assigned to an ECOD T-group |
| Proteins (cluster representatives, NCBI accessions) | 10,361, mapped to 10,109 UniProt entries |
| Domains on potentially pathogenic proteins | 3,760 (2,430 proteins) |

These are the counts of the database the website served. The paper reports
17,238 assigned domains. The 7,995 domains DPAM could not assign (probability
< 0.85) were not loaded into the database and are not part of this record.

## Structures

`ecod_salty.models_ncbi.pdb.tar.zst` holds the 10,361 full-length models the
domains were cut from, one per representative NCBI protein, as kept by the
first author for the website (`pdbs_ncbi/<NCBI acc>.pdb`). Domain ranges refer
to these models. RefSeq `WP_` accessions are written without the underscore in
file names (`WP000016413.pdb` for `WP_000016413`).

- 6,809 models carry an AlphaFold DB header (AlphaFold v2.0 monomer, AFDB
  UniProt entry named in `TITLE`) and pLDDT in the B-factor column.
- 3,552 models have no header and all B-factors 0, so pLDDT is not available
  for them. All have the length of the NCBI protein (3,551 of 3,552), i.e. they
  model the NCBI sequence. 2,541 of them differ in sequence from every version of
  the mapped UniProt entry (checked against UniParc, 2026-10-23), so they are new
  AlphaFold predictions (the paper reports 1,880 new predictions; the files do not
  explain the difference). The other 1,011 have exactly the UniProt sequence,
  so the files cannot tell whether they are new predictions or AFDB models with
  the header and pLDDT removed. The original authors no longer have that record.

`ecod_salty.models.tsv` lists every model with its UniProt accession, length,
whether it has an AFDB header, its pLDDT range, and its source (`afdb`,
`new_prediction`, `new_or_stripped_afdb`) as worked out above. The website also kept a
copy of these models named by UniProt accession (one per UniProt entry, 10,109
files). It held no extra models and is not included.

## Files

| File | Contents |
|---|---|
| `ecod_salty.domains.tsv` | **Main table.** One row per domain, with NCBI and UniProt accessions, protein annotation, residue range, pathogenicity flag and full ECOD lineage (columns below). |
| `ecod_salty.models_ncbi.pdb.tar.zst` | Full-length AlphaFold models, one PDB file per NCBI protein (see Structures). |
| `ecod_salty.models.tsv` | Per-model table: `ncbi_acc`, `model_file` (path in the archive), `unp_acc`, `residues`, `afdb_header` (`t`/`f`), `plddt_min`, `plddt_max` (empty without pLDDT), `source` (`afdb`, `new_prediction`, `new_or_stripped_afdb`). |
| `salty_dpam_domains.ncbi.txt` | DPAM output for the 17,234 domains as used for the website: ranges before small gaps were closed, DPAM and HHsearch probabilities, coverage, judgement, `Potentially_pathogenic`. |
| `salty_ncbi_uniprot_mapping.txt` | NCBI protein → UniProt accession, pathogenicity class (`Path`/`Non-Path`), protein length. No header. |
| `salty_spi_proteins_dpam_domains.ncbi.txt` | DPAM domains of the proteins in *Salmonella* pathogenicity islands, same columns as the DPAM file without the last one. No header. |
| `ecod_salty.tables.tar.zst` | Every table of the database as TSV with a header row (`domain`, `cluster`, `cluster_relation`, `unp_info`, `reference_domain`, `info`). |
| `ecod_salty.pgdump` | `pg_dump -Fc` of the PostgreSQL database (PostgreSQL 17). Restore with `pg_restore -d <db> ecod_salty.pgdump`. |
| `MANIFEST.tsv` | Size and MD5 of every file. |

Unpack the `.tar.zst` with `tar --zstd -xf` (or `zstd -dc FILE | tar -x`).

### `ecod_salty.domains.tsv` columns

Tab-separated, UTF-8, one header row, no quoting. Empty fields are missing values.

| Column | Meaning |
|---|---|
| `uid` | ECOD domain uid (9 digits, zero-padded) |
| `domain_id` | Domain identifier, `<NCBI protein acc>_nD<n>` |
| `ncbi_acc` | NCBI protein accession of the cluster representative |
| `unp_acc`, `unp_name`, `gene_name`, `protein_name` | Matching UniProt accession, entry name, gene name and protein name (names empty where UniProt has none) |
| `range` | Domain residue range on the NCBI protein's model (`ecod_salty.models_ncbi.pdb.tar.zst`); segments separated by `,`. Gaps of up to 16 residues between segments are closed; 447 domains differ from `salty_dpam_domains.ncbi.txt` only in that |
| `potentially_pathogenic` | `t` if the protein's cluster is over-represented in clinical isolates |
| `a_id`, `x_id`, `h_id`, `t_id` | ECOD architecture, X-group, H-group and T-group ids |
| `a_name`, `x_name`, `h_name`, `t_name` | Names of those groups |
| `rep_ecod_uid`, `rep_ecod_domain_id` | ECOD reference domain behind the T-group assignment |

## License and attribution

CC BY 4.0. Part of the classification is based on AlphaFold Database models
(Jumper et al. 2021, *Nature* 596:583; Varadi et al. 2022, *NAR* 50:D439).

When using these data please cite the Sci Rep paper above and ECOD:
Schaeffer RD, Liao Y, Cheng H, Grishin NV (2017). ECOD: new developments in the
evolutionary classification of domains. *Nucleic Acids Res* 45:D296–D302.
https://doi.org/10.1093/nar/gkw1137

Contact: Grishin lab, UT Southwestern Medical Center (http://prodata.swmed.edu/ecod/).