STRC Research

log

9 min read

h26 log

2026-04-27 (Phase 1d AF3 final — A1078C FAIL; h26 B → C)

  • A1078C retrieved from AlphaFold Server: fold /fold/27162e77fc95333e, archive saved as models/fold_engineered_homodimer_A1078C_afs.zip. Best model ranking_score 0.34, ipTM 0.25, pTM 0.39; all five A1078C models fail G1 (ipTM >= 0.50 and pTM >= 0.60). Final Phase 1d set is now 4/4 FAIL: A1078C 0.25/0.39, S1080C 0.33/0.41, S1579C 0.23/0.39, A1078W negative control 0.23/0.37. No G2/G4/G5/G9 follow-up justified. Ranking delta: tier B → C | mech 2 held | deliv 5 held | misha_fit 4 held | next_step → paused; no further compute unless a new scaffold/interface appears. → STRC h26 Phase 1d AF3 Final 2026-04-27

2026-04-27 (B-tier promotion triage)

  • B-tier promotion triage: h26 is the only fast possible B-to-S candidate, but only through the still-missing A1078C Phase 1d AF3 result. Local search found no A1078C ZIP under Downloads/STRC; 3/4 preliminary jobs already fail G1. Tier B held; next_step sharpened to retrieve/finish A1078C, then either demote to C on FAIL or parse G2/G4/G5 before any h03 drop-in/wetlab claim. → STRC B-Tier Promotion Triage 2026-04-27

2026-04-25 (late night — Phase 1d AF3 PRELIMINARY results: 3/4 FAIL; A1078C still running)

  • Phase 1d AF3 preliminary (3/4 jobs back, A1078C still in progress as of 23:30; AFS wall-time ~10-15 min for 3 jobs is 8-10× faster than UI estimate of “~2 hours”):
MutantipTMpTMG1 (≥0.50/0.60)URL
S1080C0.330.41🔴 FAIL/fold/73b2ae8798b7a3fe
S1579C0.230.39🔴 FAIL/fold/69aa00e31b6c2c6f
A1078W (neg-ctrl)0.230.37🔴 FAIL (expected)/fold/5403b988473e6dca
A1078C——⏳ pending—
  • Preliminary verdict: 3/3 tested disulfide-engineering candidates FAIL Gate G1 (ipTM ≥ 0.50 AND pTM ≥ 0.60). S1080C marginal best at ipTM 0.33 (vs WT 0.28-0.30, Δ +0.05) but far below threshold. Gate G2 (homodimer interface preservation) cannot be evaluated rigorously without contact-count CIF parsing, but ipTM at 0.23-0.33 implies disrupted dimer interface. A1078W negative control behaves as expected at ipTM 0.23.
  • A1078C still running at 28+ min (~3× longer than peers). Possibly genuinely harder for AF3 to converge, or AFS retry/error. Will check periodically.
  • Promotion path matrix: per hub next-step tree, “Phase 1d AF3 FAIL on all → C-tier (disulfide branch killed; no further compute)“. 3/3 confirmed FAIL; if A1078C also FAILS → h26 demote B → C + close engineered-disulfide branch.
  • Ranking delta: tier B HELD pending A1078C | mech 2 held | deliv 5 held | misha_fit 4 held | next_step held until A1078C lands. Gates G1+G2 = preliminary FAIL; G3 PASS analytical (Phase 1c) unaffected; G4+G5+G9 NOT RUN (deferred until any candidate passes G1+G2). → STRC h26 Phase 1d AF3 Preliminary 2026-04-25 (note to be written when A1078C lands).

2026-04-25 (late night — gate-discipline pilot)

  • Methodology import: added explicit gates: block to h26 hub frontmatter + ## Gates section in body, adapting PeptAI’s per-stage threshold pattern to h26’s structural objective. G1 (AF3 quality) + G2 (interface preservation) + G3 (disulfide geometry) + G4 (interface energy) + G5 (MD stability) + G9 (h03 drop-in compatibility) carried over; G0 ChEMBL baseline merged into G2 (WT contact count + A1078W negative control); G6–G8 ADMET dropped (h26 is intracellular AAV-payload protein, not a systemic peptide drug). G3 already PASS analytical from Phase 1c; G1+G2 PENDING Phase 1d AF3; G4+G5+G9 NOT RUN. Ranking delta: tier B held | mech 2 held | deliv 5 held | misha_fit 4 held | next_step held (Phase 1d AF3). No proof, no new compute, no script — discipline overlay only. → h26 Gate Pilot — PeptAI Discipline Import 2026-04-25

2026-04-25 (late night — Phase 1d AF3 batch SUBMITTED to AlphaFold Server)

  • Phase 1d AF3 jobs submitted via Playwright MCP. 4 jobs in ~/STRC/models/af3_jobs_2026-04-25_h26_phase1d/: A1078C / S1080C / S1579C homotypic homodimer mutants + A1078W negative control. Construct = Ultra-Mini × 2 + TMEM145 × 2 (matches Phase 1c reference template). Mutations applied at Mini-STRC index 4 (A1078), 6 (S1080), 505 (S1579). All 4 passed JSON upload validation (check_circle, no errors), all 4 promoted from draft to running via 3-click flow per ~/.claude/skills/alphafold-server/SKILL.md Step 1b.
  • Process discovery: AFS “Submit N jobs as drafts” button only saves drafts; explicit row → Continue → Confirm cycle required to actually launch. Documented in alphafold-server skill Step 1b. Single-file input forces JSON-array packing for batch upload — also documented.
  • Ranking delta: tier B held | mech 2 held | deliv 5 held | misha_fit 4 held. Decision pending Phase 1d AF3 results (~6-12 h AFS queue). Promotion path remains: any mutant ipTM ≥ 0.50 + homodimer contacts ≥ WT → A-tier → drop into h03 Ultra-Mini as engineered-avidity variant (zero payload cost). → STRC h26 Phase 1d AF3 Submission 2026-04-25 (note to be written when results land).

2026-04-25 (night — literature ingest: Schneider Ed. 2014 De novo Molecular Design)

  • Literature: ingested De novo Molecular Design (Schneider Ed., Wiley-VCH 2014). Source 2014-schneider-baringhaus-de-novo-molecular-design-book; paper note 2014-schneider-de-novo-molecular-design-book. Ch. 19 (Saven, computational protein design) provides method survey for the cysteine-engineering Phase 1d work — DEE / SCMF / probabilistic side-chain search is the literature-backed protocol class for triple-mutant repacking. Ch. 17 (bioisosteres) provides fall-back replacement strategies if A1078C / S1080C / S1579C disulfides do not stabilize. The cross-cutting Fragment Additivity Assumption and Superadditivity concept warns that disulfide ΔΔG_dimer cannot be summed from per-residue mutation scores — must run FEP point-mutation on the assembled dimer. Distance matrix in Amino Acid Physicochemical Distance Matrix Grantham Modified confirms cysteine substitution is the worst case for SME-style continuous evolution (column-C distances ≈ 1) — explicit design (Rosetta DESIGN, AF3 disulfide-finder) remains the right tool. Ranking delta: no tier/mech/deliv/misha_fit/next_step change. Tier B held; methods foundation strengthened for upcoming Phase 1d.

2026-04-25

2026-04-23

  • Phase 1c contact re-cluster: GREEN. DBSCAN eps=6.5 Å on Ultra-Mini homodimer CIF found 3 clusters. Cluster 1 “stump” 1077-1114 (25 residues, 21 Cb-Cb disulfide geometries) ENTIRELY outside Phase 1 target 1579-1581. Native C1079+C1081 on both chains sit 6.87-7.09 Å from A1078/S1080 on opposing chain → single A1078C or S1080C engineers inter-chain disulfide without changing cystine count. Cluster 2 ARM has homotypic S1579C/S1579C at Cb-Cb 6.94 Å (re-opens Phase 1 target with disulfide chemistry). B held pending Phase 1d AF3 triple batch (A1078C + S1080C + S1579C homotypic + A1078W negative control). Promotion path: Phase 1d ipTM ≥0.50 + homodimer contacts ≥ WT → A-tier → drop into h03 Ultra-Mini as engineered-avidity variant (zero payload cost). → STRC h26 Phase 1c Contact Re-Cluster 2026-04-23
  • Added to register: A-tier, Mech 4, Deliv 5, Misha 4. R-R repulsion at ARM 1579-1581 identified as dimer destabilizer → STRC Engineered Homodimer Avidity
  • Phase 1 AF3 batch built: 5 jobs (WT + 4 single-point mutants × 2 TMEM145), awaiting submission → STRC Engineered Homodimer Avidity
  • Phase 1 AF3 returned 4 mutants (WT_ref not downloaded): all FAIL homodimer gate (mean 0.15-0.19 vs WT 0.28-0.30); 3/4 also destroy TMEM145 binding → STRC Engineered Homodimer Phase 1 Results
  • Tier A → B. Mech 4→3. R-R repulsion falsified by R1581F (arg removed, homodimer still weak) → STRC Engineered Homodimer Phase 1 Results
  • Dunbrack 2025 ipSAE reassessment: Ultra-Mini homodimer ipSAE 0.000 (zero interface residues pass PAE<10 Å cutoff at default AF3 settings) — ipTM 0.28-0.30 uplift from baseline 0.20-0.24 now identified as disordered-region false-positive pattern ipSAE was designed to catch. Mech 3→2, B held. Structural symmetry (94% C2) + aa 1579-1581 self-contacts legs survive the reassessment (they are geometry observations independent of the confidence metric). Phase 1c 5 Å re-cluster still the next viable move. → STRC ipSAE Cross-Complex Reassessment 2026-04-23

2026-04-23

  • #26 Engineered Homodimer Phase 1c GREEN: spatial DBSCAN (eps 6.5 Å) re-cluster of Ultra-Mini homodimer inter-chain contacts found 3 clusters. Cluster 1 “stump” 1077-1114 (25 residues, 21 disulfide-geometry Cb-Cb in 4.5-7.5 Å window) is entirely outside Phase 1’s tested 1579-1581 target. Native C1081 sits within Cb-Cb 6.87-7.09 Å of A1078/C1079/S1080 on opposing chain — latent inter-chain cystine site. Cluster 2 (ARM 1529-1590) also contains untouched candidates incl. homotypic S1579→C/S1579→C symmetric disulfide at Cb-Cb 6.94 Å (re-opens ARM target with different chemistry than Phase 1’s aromatic Y/F mutants). B held pending Phase 1d AF3 triple batch (A1078C, S1080C, S1579C homotypic mutants + A1078W negative control). Promotion path: Phase 1d ipTM ≥ 0.50 + homodimer contacts ≥ WT → A-tier → drop-in to h03 Ultra-Mini as engineered-avidity variant. → STRC h26 Phase 1c Contact Re-Cluster 2026-04-23
  • #9 Hydrogel + #26 Homodimer: ipSAE (Dunbrack 2025) reassessment on 6 STRC AF3 jobs (incl. 2 known positive + 2 known negative controls). Ultra-Mini × TMEM145 GOLD ipSAE 0.591 sits in known-binder-zone (NFAT-CnB 0.55 … NFAT-CnA 0.78 physiological binders) → h09 gate 3 narrowed from “unsupported placeholder” to “band-consistent 10 nM-10 µM, wet-lab still gating absolute Kd”; A held, Mech 4 held. Ultra-Mini homodimer ipSAE 0.000 (zero interface residues pass PAE<10 Å) contradicts prior ipTM 0.28-0.30 weak-binder interpretation per Dunbrack’s disordered-region-false-positive fix → #26 Mech 3→2, B held. Path B (AF3 absolute Kd calibration) closed as lit-infeasible. → STRC ipSAE Cross-Complex Reassessment 2026-04-23
  • #9 Hydrogel + #26 Homodimer: ipSAE (Dunbrack 2025) reassessment on 6 STRC AF3 jobs (incl. 2 known positive + 2 known negative controls). Ultra-Mini × TMEM145 GOLD ipSAE 0.591 sits in known-binder-zone (NFAT-CnB 0.55 … NFAT-CnA 0.78 physiological binders) → h09 gate 3 narrowed from “unsupported placeholder” to “band-consistent 10 nM-10 µM, wet-lab still gating absolute Kd”; A held, Mech 4 held. Ultra-Mini homodimer ipSAE 0.000 (zero interface residues pass PAE<10 Å) contradicts prior ipTM 0.28-0.30 weak-binder interpretation per Dunbrack’s disordered-region-false-positive fix → #26 Mech 3→2, B held. Path B (AF3 absolute Kd calibration) closed as lit-infeasible. → STRC ipSAE Cross-Complex Reassessment 2026-04-23
  • #26 Engineered Homodimer: A → B. Mech 4→3. Phase 1 AF3 ALL FAIL: single-point mutants destabilize homodimer (mean 0.15-0.19 vs WT 0.28-0.30); R-R repulsion hypothesis falsified. → STRC Engineered Homodimer Phase 1 Results
  • #26 STRC Engineered Homodimer Avidity added, A-tier. Phase 1 AF3 batch built (5 jobs). → STRC Engineered Homodimer Avidity
  • #26 Homodimer Avidity post-audit fix: B held. Prose phantoms reframed: “100 nM → 1 nM” Kd claims now explicit ratio-on-unmeasured-baseline; status banner notes Phase 1 AF3 fail + unmeasured Kd; avidity principle (Jencks 1981, Mammen 1998, Kramer & Karpen 1998) real; absolute Kd claims placeholder. Mech 4→3 (Phase 1 structural fail). → STRC Engineered Homodimer Avidity

Connections

Graph View