Loading H-group summaries — first cold-cache hit can take a few seconds.
Paper Fig 5A,B
H-groups binned by structurally-known cysteine fraction (PDB-source representatives) versus ESM2-predicted cysteine fraction (all F70 representatives in the H-group). Cells in the lower-right quadrant — low structural evidence, high ESM2 prediction — flag candidate novel cysteine-chemistry H-groups; the manuscript highlights 12 disulfide and 3 metal-binding such cells.
The three H-groups featured in Fig 5C–E exemplify the low-known / high-predicted cell of the metal-binding confusion matrix.
Candidate novel metal-binding H-group: PDB-source representatives carry no metal-LINK evidence, but ESM2-3state predicts metal-binding cysteines at high confidence in matched AFDB representatives.
Candidate novel metal-binding H-group with sub-threshold PDB-source evidence and high ESM2-3state metal-binding rates across AFDB-source representatives.
Candidate novel metal-binding H-group: third example illustrating the "low PDB known, high ESM2 predicted" cell of the Fig 5 confusion matrix.
Fig 5A · Disulfide
Click any cell to list the H-groups that fall into it. The amber-ringed cell flags the candidate novel quadrant.
| ESM2-predicted (AFDB-source) | ||||
|---|---|---|---|---|
| <5% | 5–50% | 50–95% | ≥95% | |
| Structurally known (PDB-source) | ||||
| <5% | 5–50% | 50–95% | ≥95% | |
Showing 3,352 H-groups with at least one F70 representative. H-groups with no PDB-source representative fall outside the matrix axis (no structurally-known signal) and are excluded from cell counts; those with no AFDB-source representative are similarly excluded.
Fig 5B · Metal-binding
The Fig 5C–E highlighted H-groups live in the lower-right cell of this matrix.
| ESM2-predicted (AFDB-source) | ||||
|---|---|---|---|---|
| <5% | 5–50% | 50–95% | ≥95% | |
| Structurally known (PDB-source) | ||||
| <5% | 5–50% | 50–95% | ≥95% | |