STRC h09 Phase 4b tail91_v2 developability scrub
tail91_v2 Cys11/52→Ser keeps TMEM145 and actin axes robust; deferred seed777/triple checks do not reverse the production-candidate verdict.
Problem
tail91_v2 removes the two free cysteines from the h09 RADA16-WH2-tail91 construct. The developability question was whether Cys11/52→Ser keeps both binding axes robust while avoiding oxidation/disulfide-shuffling liabilities.
Method
AlphaFold Server jobs from ~/STRC/models/af3_jobs_2026-04-27_h09_phase4b_tail91_v2/ were read from fold-detail ipTM/pTM pages and 2026-04-27 downloaded ZIP summary_confidences_0.json files.
Result
Artifact: hypotheses/h09-hydrogel/artifacts/phase4b_tail91_v2_2026-04-26_results.json.
| Axis | n | ipTM median | sigma | Verdict |
|---|---|---|---|---|
| TMEM145 | 5 | 0.55 | 0.010 | PASS |
| Actin | 5 | 0.51 | 0.011 | PASS |
Deferred checks completed on 2026-04-27:
| Job | ipTM | pTM | Interpretation |
|---|---|---|---|
hydrogel_tail91_v2_x_actin_seed777 | 0.51 | 0.63 | Confirms actin axis remains at threshold across all 5 seeds. |
hydrogel_tail91_v2_x_tmem145_x_actin | 0.34 | 0.46 | Cross-bind topology unchanged vs v0 triple ipTM 0.32; not a reversal. |
Interpretation
The scrub is not completely binding-neutral: TMEM145 median drops from 0.60 to 0.55. That is a real reproducible cost, but the axis remains above the 0.50 robustness gate with tighter variance than v0.
The actin axis is unchanged after the deferred seed777 read: 5-seed median 0.51, sigma 0.011. The v2 triple-complex does not solve cross-binding, but it also does not worsen it.
Verdict
- Outcome: PASS.
- Key numbers: TMEM145 median ipTM 0.55, actin median ipTM 0.51, v2 triple ipTM 0.34.
- Limitations: AF3 ipTM is a structural-confidence proxy, not a Kd measurement. The WH2 x F-actin Kd-equivalent gate remains open.
Ranking delta
- Tier: tentative-S → tentative-S. No demotion trigger from seed777 or v2 triple.
- Mech: 5 → 5. Developability scrub preserves both single-axis AF3 robustness gates.
- Deliv: 4 → 4.
- Misha-fit: 3 → 3.
- Next step: unchanged. Full S still depends on closing the WH2 x F-actin Kd-equivalent uncertainty via Phase 4m, 4l, or confidence-supported Phase 4k.