# cath_v4_5_0.prerelease.homcheck_s95_cross_assignments.tsv

Preliminary. For the CATH v4.5.0 prerelease. Rows may change if the HomCheck gates change before release.

## What this data represents

S95 domains that HomCheck assigns to a superfamily other than the CATH v4.4 label. A row is included when Contrasted and Foldseek both pass and name the same superfamily. TEDLH is on the row. It is not required for inclusion.

Tab-separated. Header on line 1. One domain per row. Labels are CATH v4.4.

## How the data was generated

S95 leave-one-out, 69,592 domains (classes 1-3, superfamily size greater than 1). Each domain is scored against the other s95 domains.

Contrasted uses the rank-1 neighbour in the projected 128-d AA+3Di embedding. It passes when 1-cosine <= 0.25 and min(Lq, Lt)/max(Lq, Lt) >= 0.60.

Foldseek uses the rank-1 3Di+AA hit. It passes when E <= 0.01, native TM >= 0.50, and native coverage >= 0.60. Native TM is min(qTM, tTM). Native coverage is min(qcov, tcov).

TEDLH is the top hit in its published box: probability >= 90, E <= 0.001, query coverage >= 0.70, target coverage >= 0.70. `tedlh` is `same` when that superfamily equals `assigned_superfamily`, `different` when it is another superfamily, and `abstain` when no hit passes the box. `tedlh_superfamily` is that superfamily, or blank when TEDLH abstains.

`seq_identity` is the CATH sequence identity of `domain` against `foldseek_domain`. The sequences are aligned with Needleman-Wunsch, scoring 1 for an identical residue and 0 otherwise, with a gap penalty of 4. Identity is the number of identical residues divided by the length of the smaller domain. The value is a fraction from 0 to 1. CATH quotes the same quantity as a percentage.

## Columns

| Column | Meaning |
|---|---|
| domain | Query domain |
| superfamily | v4.4 superfamily |
| assigned_superfamily | Superfamily named by Contrasted and Foldseek |
| confidence | `high`: all three methods assign that superfamily. `medium`: Contrasted and Foldseek agree, TEDLH abstains. `two_vs_one`: Contrasted and Foldseek agree, TEDLH assigns a different superfamily |
| contrasted_domain | Contrasted rank-1 neighbour |
| contrasted_distance | 1-cosine to `contrasted_domain` |
| foldseek_domain | Foldseek rank-1 hit |
| foldseek_evalue | Foldseek E-value |
| foldseek_tm | Foldseek native TM |
| foldseek_coverage | Foldseek native coverage |
| seq_identity | Identical residues divided by the length of the smaller domain, for `domain` vs `foldseek_domain` |
| tedlh | `same`, `different`, or `abstain` |
| tedlh_superfamily | Superfamily TEDLH assigns. Blank when `tedlh` is `abstain` |

## Shape

60 rows. 60 domains. 43 superfamily pairs. 51 labelled superfamilies.

`confidence`: `high` 37, `medium` 6, `two_vs_one` 17. These match `tedlh` `same`, `abstain`, and `different`.

Curator review of 40 of these domains: `cath_v4_5_0.prerelease.homcheck_s95_curator_judgements.tsv`.