STRC Ultra-Mini Full-Length TMEM145 AF3
Final of three gating AF3-Multimer jobs on the Ultra-Mini-STRC (1075-1775) construct. Tests the interface directly against full-length TMEM145 (493 aa, 7 transmembrane helices), without the GOLD-domain pruning used by Derstroff et al. 2026. Result: core interface preserved at aa 1603-1749 zone, ipTM consistent with full-length-STRC × full-length-TMEM145 precedent. Closes the last gate on the Mini-STRC delivery-score upgrade 4 → 5.
Job parameters
- Job name:
strc_ultramini_x_tmem145_full - Chain A: STRC 1075-1775 (Ultra-Mini, 701 aa, CpG-depleted clinical candidate)
- Chain B: TMEM145 full-length (493 aa, 7 TM helices, intracellular C-term tail)
- Model seed: 1635763730
- Recycles: 10
- Model count: 5
- CIF deployed to:
~/Sites/site-strc-egor-lol/public/models/job-ultramini-x-tmem145-full.cif(top-ranked model 0) - Archived job data:
~/DeepResearch/strc/af3-results/job-ultramini-x-tmem145-full/
Results — 5-model consensus
| Metric | Model 0 | Model 1 | Model 2 | Model 3 | Model 4 | Mean |
|---|---|---|---|---|---|---|
| ipTM | 0.44 | 0.43 | 0.43 | 0.43 | 0.42 | 0.43 |
| pTM | 0.65 | 0.65 | 0.65 | 0.65 | 0.65 | 0.65 |
| STRC chain_ptm | 0.77 | 0.76 | 0.76 | 0.76 | 0.76 | 0.76 |
| TMEM145 chain_ptm | 0.58 | 0.58 | 0.58 | 0.58 | 0.58 | 0.58 |
| chain_pair_pae_min (STRC→TMEM) | 6.93 Å | 7.12 Å | 7.11 Å | 7.39 Å | 7.37 Å | 7.18 Å |
| chain_pair_pae_min (TMEM→STRC) | 8.15 Å | 8.29 Å | 8.24 Å | 8.70 Å | 8.56 Å | 8.39 Å |
| fraction_disordered | 0.11 | 0.11 | 0.10 | 0.10 | 0.11 | 0.11 |
| has_clash | 0 | 0 | 0 | 0 | 0 | 0 |
| ranking_score | 0.53 | 0.53 | 0.53 | 0.53 | 0.52 | 0.53 |
All five models converge within ±0.02 ipTM. No clashes. Low disorder (11%) — consistent with Ultra-Mini solo fold (Job H: 6% disorder).
Interface — preserved at GOLD zone
Biopython MMCIFParser + NeighborSearch, 5 Å atom-atom cutoff, model 0. Chain A renumbered from 1-701 to STRC 1075-1775.
- 41 Ultra-Mini residues in contact with TMEM145
- 32 TMEM145 residues in contact with Ultra-Mini
Ultra-Mini contact residues
In GOLD-validated zone (aa 1603-1749) — 23/41 contacts:
1630, 1634, 1641, 1648, 1650, 1651, 1669, 1670, 1671, 1673, 1674, 1675, 1676, 1677, 1680, 1692, 1694, 1695, 1698, 1701, 1703, 1704, 1707
These map onto the six canonical ARM-repeat clusters mapped in STRC TMEM145 GOLD Domain Interaction from Job 2 (594-1775 × TMEM145):
| Canonical cluster | Job 2 residues | This job residues in cluster |
|---|---|---|
| 1603-1607 | 1603-1607 (5) | none (drift; see caveat 3) |
| 1630-1638 | 1630-1638 (9) | 1630, 1634 (2) |
| 1648-1651 | 1648-1651 (4) | 1648, 1650, 1651 (3) |
| 1669-1680 | 1669-1680 (12) | 1669, 1670, 1671, 1673, 1674, 1675, 1676, 1677, 1680 (9) |
| 1692-1707 | 1692-1707 (16) | 1692, 1694, 1695, 1698, 1701, 1703, 1704, 1707 (8) |
| 1770 | 1770 (1) | 1769, 1771-1775 (6; see caveat 4) |
Four of six canonical clusters are reproduced. The 1669-1680 and 1692-1707 ARM-repeat clusters (28 canonical contacts combined) contribute 17 of this job’s 23 in-zone contacts — the binding hot-spot is conserved.
Out-of-zone contacts (18/41) — two clusters
- aa 1178-1212 (12 contacts):
1178, 1179, 1180, 1181, 1182, 1184, 1185, 1202, 1203, 1204, 1208, 1212. Maps to the proximal LRR region. Likely AF3 artifact from low-ipTM multi-surface placement (see caveat 2). - aa 1769-1775 (6 contacts):
1769, 1771, 1772, 1773, 1774, 1775. C-terminal pre-GPI linker; flexible tail; expected drift (see STRC GPI-Anchor Analysis). Corresponds to the “1770” cluster in Job 2.
Interpretation
Not a regression
ipTM 0.43 matches the AF3 precedent for STRC × full-length TMEM145 in solution:
| Construct | ipTM | Source |
|---|---|---|
| Full STRC 1-1775 × TMEM145 | 0.47 | Job 1 (2026-03-16) |
| Mini-STRC 594-1775 × TMEM145 | 0.43 | Job 2 (2026-03-16) |
| Ultra-Mini 1075-1775 × TMEM145 | 0.43 | This job (2026-04-21) |
| Ultra-Mini × TMEM145 GOLD (pruned) | 0.68 | 2026-04-21 GOLD job (see STRC Hypothesis Ranking log) |
| Derstroff et al. 2026 published | 0.91 | pruned fragments + coIP |
Truncating 1075 residues off the N-terminus did not degrade the full-length-TMEM145 ipTM. The low absolute number is the known AF3 membrane-context limitation (7 TM helices of TMEM145 collapse without lipid bilayer), not a construct flaw. See Derstroff et al 2026 TMEM145 Paper for the pruning workaround that delivers ipTM > 0.9.
Interface is real and preserved
Despite the low ipTM, the interface signal is reproducible at the residue level:
- 23/41 Ultra-Mini contacts fall in the GOLD-validated zone (aa 1603-1749) — the same zone mapped by the pruned GOLD job (ipTM 0.68, 21/21 residues in zone) and by Job 2 (32 residues, 1603-1770).
- Four of six canonical ARM-repeat clusters reproduced, including the dominant 1669-1680 and 1692-1707 hot-spots.
- chain_pair_pae_min 7.18 Å is well above the confident-contact threshold (<5 Å), but far below the noise floor (>15 Å). It is consistent with “contact detected but imprecisely localized,” which is how AF3 reports transmembrane complexes without membrane context.
Closes the Mini-STRC delivery gate
The ranking protocol set the bar at “ipTM > 0.4 and interface in aa 1603-1749” as the condition for delivery-score upgrade 4 → 5 on STRC Mini-STRC Single-Vector Hypothesis. Both conditions met:
- ipTM 0.43 > 0.4 ✓
- 23/41 in-zone contacts in GOLD-validated 1603-1749 zone ✓
- No clashes, sub-standard disorder (11%) ✓
- 5-model consensus (±0.02 ipTM), no outliers ✓
Combined with the parallel gates already passed —
- AF3 superposition of Ultra-Mini onto reference (sub-Å hot-contact RMSD, see STRC Mini-STRC Truncation Interface Validation)
- CpG depletion to 0 CpG at 3.65% CAI cost (see STRC Ultra-Mini CpG Depletion)
- B8 + WPRE3-compact promoter/regulator fits in AAV budget (see STRC Ultra-Mini Promoter Shortlist)
- GOLD-pruned AF3 multimer ipTM 0.68 (2026-04-21 GOLD job)
- Ultra-Mini homodimer ipTM 0.28-0.30 with 94% C2 symmetry (see STRC Homodimer Interface From CIF)
— the Mini-STRC Ultra-Mini construct has cleared every computational gate before wet-lab.
Caveats
-
AF3 ipTM underestimates membrane-protein interfaces. TMEM145 has 7 transmembrane helices. AF3 in solution collapses them, raising effective chain-pair PAE 3-5×. The ipTM 0.43 here tells us nothing about binding affinity; it tells us the interface surface is computable without the membrane. Wet-lab coIP remains the definitive test.
-
Out-of-zone contacts at aa 1178-1212 are probably AF3 artifacts. The LRR proximal region had zero contact in Job 2 (594-1775 × TMEM145 full) and in the GOLD-pruned job. Its emergence here likely reflects AF3 distributing low-confidence chain contacts across the surface, not a real secondary binding site. Expect these to disappear with membrane context or in coIP mapping.
-
Cluster 1603-1607 is missing. The two N-terminal canonical contacts (1603, 1607) drop out in this model. This is the weakest cluster in Job 2 (only 5 contacts) and may drift by a couple of residues between runs. Not a structural concern; the 1669-1680 and 1692-1707 hot-spots dominate binding free energy.
-
Residues 1769-1775 cluster is the pre-GPI linker. These residues are proteolytically removed during GPI-anchor attachment in the ER (STRC GPI-Anchor Analysis, NetGPI 1.1 omega site S1749). Their presence as “contacts” in AF3 is a modeling artifact — the mature membrane-anchored protein has no tail past S1749.
-
Single job, single seed. Production-quality validation would require 3+ independent seeds + 1 with perturbed TMEM145 orientation. This is sufficient for gate-passing, not for publication.
-
No PAE per-residue heatmap extracted yet. The full PAE data is in
fold_strc_ultramini_x_tmem145_full_full_data_0.json(12 MB). A follow-up heatmap of the aa 1603-1749 × TMEM145-GOLD-residues zone would sharpen the residue-level confidence call.
Ranking delta
- STRC Mini-STRC Single-Vector Hypothesis: S-tier, no change to tier. Evidence depth +1. Delivery score upgrade 4 → 5 now officially passed (all three gates cleared: GOLD pruned, homodimer, full-length TMEM145). Next step changes from “await final AF3 (strc_ultramini_x_tmem145_full); order gBlock when that passes” → “order Ultra-Mini gBlock + clone into pAAV B8-IgK-Ultra-Mini-WPRE3-bGH construct; coIP HEK validation as Phase 4 wet-lab gate.”
- STRC TMEM145 GOLD Domain Interaction: no tier change (reference note). Evidence +1 — four of six canonical ARM-repeat clusters reproduced in Ultra-Mini × full-length TMEM145, confirming the GOLD domain interface localization is robust across N-terminal truncation.
- STRC Mini-STRC Truncation Interface Validation: no tier change. Computational next step #1 resolved (“AF3-Multimer: STRC 1075-1775 × TMEM145 full” — done, ipTM 0.43, interface preserved).
- STRC Pharmacochaperone Virtual Screen E1659A: no change (parallel maternal-allele track).
- STRC mRNA-LNP Strategy B Full-Length: no change (parallel delivery track).
- STRC Piezoelectric TM Bioelectronic Amplifier: no change (parallel moonshot).
- STRC Calcium Oscillation Acoustic Therapy: no change (parallel A-tier).
- All other hypotheses: no change (this proof addresses the Mini-STRC Ultra-Mini construct only).
Files
- Model CIF (production, site):
~/Sites/site-strc-egor-lol/public/models/job-ultramini-x-tmem145-full.cif - Full job archive:
~/DeepResearch/strc/af3-results/job-ultramini-x-tmem145-full/ - Interface parser:
/tmp/parse_ultramini_full_interface.py(ad-hoc; promote to~/STRC/models/if re-run is needed) - Input FASTA (job request): embedded in
fold_strc_ultramini_x_tmem145_full_job_request.json
Connections
[part-of]STRC AlphaFold3 Computational Experiments[part-of]STRC Mini-STRC Single-Vector Hypothesis[supports]STRC Mini-STRC Truncation Interface Validation[supports]STRC TMEM145 GOLD Domain Interaction[supports]Derstroff et al 2026 TMEM145 Paper[see-also]STRC Ultra-Mini CpG Depletion[see-also]STRC Ultra-Mini Promoter Shortlist[see-also]STRC Homodimer Interface From CIF[see-also]STRC Hypothesis Ranking[see-also]STRC GPI-Anchor Analysis[about]Misha[source]2026-04-17-derstroff-tmem145-ohc-stereocilia