Scope (narrowed 2026-04-15)
Originally filed with 82 FPs (3.8%). Recent work (h-region leaderless gate #42, charged-penalty normalization #43, SP -3 grammar fix #38, motif anchors) has brought total FPs down to 34 (1.6%) on 2155 controls.
The remaining pool is concentrated in one species category:
| Category |
FPs |
Total |
Rate |
| other_vertebrate (mature-only) |
23 |
1011 |
2.3% |
| fish |
7 |
379 |
1.8% |
| murine |
3 |
75 |
4.0% |
| bird |
1 |
305 |
0.3% |
| All others |
0 |
— |
— |
The 23 other_vertebrate FPs are almost entirely reptile non-classical class I (E-S / F10 / Q9 gene families from Pogona, Python, Salvator, Anolis, etc.) where mature-only deposits present N-terminal groove helices that mimic SP h-regions.
Approach options (unchanged)
- Groove-helix fingerprinting: groove helices have different composition than SP h-regions
- Sequence-length-aware gating: 200-280 aa without Met = likely mature-only
- N-terminal domain compatibility: well-formed groove from pos 0 = evidence against SP
Success criterion
Drive the other_vertebrate FP rate below 1% (≤10 FPs) without regressing full-length class I SP accuracy on reptile/fish.
Scope (narrowed 2026-04-15)
Originally filed with 82 FPs (3.8%). Recent work (h-region leaderless gate #42, charged-penalty normalization #43, SP -3 grammar fix #38, motif anchors) has brought total FPs down to 34 (1.6%) on 2155 controls.
The remaining pool is concentrated in one species category:
The 23 other_vertebrate FPs are almost entirely reptile non-classical class I (E-S / F10 / Q9 gene families from Pogona, Python, Salvator, Anolis, etc.) where mature-only deposits present N-terminal groove helices that mimic SP h-regions.
Approach options (unchanged)
Success criterion
Drive the other_vertebrate FP rate below 1% (≤10 FPs) without regressing full-length class I SP accuracy on reptile/fish.