STRC h09 Phase 4m — Kd_mono soft-floor refinement (pre-registration)

Status: pre-registered. No compute consumed. Defines protocol + thresholds before running so post-hoc selection is impossible.

Goal. Soften the WH2 × F-actin Kd_mono floor used in STRC h09 Phase 4j Avidity Kd Model 2026-04-26 from the Husson 2010 Tβ4 × F-actin = 5–10 mM analog floor down to a comparative-AF3-derived bound. If 4m succeeds, Phase 4j conservative scenario gains a more favorable Kd_mono input → conservative gate (Kd_eff < 50 µM) closes.

Hypothesis

AF3 ipTM is NOT a calibrated Kd predictor (avidity-and-dimers.md §3, Dunbrack ipSAE 2025), but comparative ipTM ratios within a single benchmark family of known-Kd binders can place a soft Kd floor on a related construct of unknown Kd.

Specifically: by running AF3 on the Chereau 2005 ITC-measured WH2 × G-actin family (WAVE2, WASP, WIP, MIM, Tβ4) under matched conditions, we establish a within-family ipTM-vs-Kd relationship. Then run h09 tail91 WH2 × G-actin (same conditions) and project Kd from its ipTM rank.

Caveat baked in: even within a single family, ipTM-Kd correlation is order-of-magnitude at best. Output is a “soft Kd floor” (lower bound on plausible Kd_mono), not a point estimate.

Reference set (Chereau 2005 Fig. 1C, primary lit, all in vault as ~/STRC/literature/actin-kinetics.md)

WH2 sourceSequence positionKd × G-actin (ITC)
WAVE2433–464 (32 aa)52 ± 3 nM
WIP29–60 (32 aa)160 ± 10 nM
MIM724–755 (32 aa)230 ± 40 nM
WASP430–458 (29 aa)250 ± 30 nM
Tβ42–44 (43 aa)760 ± 120 nM (G-actin only; F-actin floor 5–10 mM cited separately)

Protocol

AF3 jobs (5 reference + 1 query = 6 total; well under daily 30 quota)

  1. WAVE2_WH2_433_464 × G-actin trimer (single seed=42)
  2. WIP_WH2_29_60 × G-actin trimer
  3. MIM_WH2_724_755 × G-actin trimer
  4. WASP_WH2_430_458 × G-actin trimer
  5. Tβ4_WH2_2_44 × G-actin trimer
  6. h09 tail91 N-term WH2 (RQLVKAIPDNCSKSNVSR, aa 1–18 of construct) × G-actin trimer

All vs G-actin trimer (the Chereau ITC measured G-actin → use G-actin as substrate to keep comparison clean). G-actin trimer = 3 × 375 aa = 1125 aa.

WH2 sequences (need to retrieve from UniProt for reference set):

  • WAVE2 (Q9Y6W5) aa 433–464: [TBD — retrieve before submission]
  • WIP (Q14155) aa 29–60: [TBD]
  • MIM (Q8WUY3) aa 724–755: [TBD]
  • WASP (P42768) aa 430–458: [TBD]
  • Tβ4 (P62328) aa 2–44: [TBD]
  • h09 WH2 (custom): RQLVKAIPDNCSKSNVSR (already known)

Token budget per job: 32 aa + 1125 aa = 1157 aa (well under 5000).

Analysis

For each reference job, record top-model ipTM. Plot log(Kd_ITC) vs ipTM. Fit linear regression. Apply h09 WH2 ipTM to the line → predicted log(Kd) for h09 WH2 × G-actin.

Then propagate: h09 Kd × F-actin = h09 Kd × G-actin × scaling_factor. Scaling factor from Tβ4 family: Tβ4 × G-actin = 760 nM, Tβ4 × F-actin = 5–10 mM → ratio ≈ 6,500–13,000×. Apply ratio to predicted h09 Kd × G-actin.

Pass criteria (pre-registered)

  1. R² of family fit ≥ 0.5. If reference family fit is too noisy, abandon the comparison; soft-floor cannot be established.
  2. h09 ipTM lies within reference range. Extrapolation outside (0.4–0.7) is unreliable; if h09 ipTM < 0.4 or > 0.75, abandon.
  3. Predicted h09 Kd × G-actin × F-actin scaling < 5 mM. This becomes the soft floor that replaces Husson Tβ4 5–10 mM in Phase 4j.
  4. If predicted h09 Kd_mono < 1 mM, conservative Phase 4j scenario closes: Kd_eff = 1e-3 × (1e-3/0.05)^3 = 8 µM (PASS gate 50 µM).

What this does NOT prove

  • Does NOT measure absolute Kd. Only places a soft floor for Phase 4j input.
  • Does NOT validate AF3 ipTM as Kd predictor in general (it isn’t). Within-family comparative ratios are the only legitimate use.
  • Does NOT close the wet-lab gate (which is off-limits per project rules); it tightens the computational substitute.

Cost

6 AF3 jobs (1 quota cycle, ~25 min wall-time). No M5 compute. Pure analytic afterward.

Status

Pre-registered 2026-04-26 16:??. Not submitted. Submit when AFS quota refreshes (~midnight UTC after retrieving WH2 reference sequences from UniProt).

Connections