STRC h09 Phase 4k Fibril Actin Geometry

Phase 4k INCONCLUSIVE: tandem proxies show 4 geometric WH2-actin contacts, but 0 confidence-supported contacts across all 5 AF3 jobs.

Problem

Phase 4k tested whether AF3 could support the Phase 4j avidity requirement that multiple WH2 motifs can contact actin in the same local fibril/filament patch. The pre-registered cheap gate was N>=4 simultaneous WH2-actin engagements in at least one construct.

Method

  • Script: hypotheses/h09-hydrogel/scripts/phase4k_af3_contact_parser.py
  • Inputs: five AlphaFold Server ZIPs in models/af3_results/h09_phase4k_4m_2026-04-27/
  • Artifact: hypotheses/h09-hydrogel/artifacts/phase4k_contact_parse_2026-04-27.json
  • Contact rule: WH2 motif heavy atom within 5.0 A of actin.
  • Confidence rule: contact plus chain-pair ipTM >= 0.20.

Results

JobipTMpTMGeometric WH2-actin contactsConfidence-supported contactsGate
construct x2 + actin x30.450.5700FAIL
construct x3 + actin x30.410.5320FAIL
construct x4 + actin x30.370.4900FAIL
tandem WH2 x4 + actin x30.570.6840geometric-only
tandem WH2 x4 + actin x50.720.7640geometric-only

The final x4 construct was slow on AFS: the original row eventually completed and was downloaded; the duplicate recovery row was still running when the original result became available.

Verdict

  • Outcome: INCONCLUSIVE
  • Key numbers: any geometric pass = true; any confidence-supported pass = false; all WH2-actin chain-pair ipTM values remain below the 0.20 confidence floor.
  • Interpretation: AF3 can place tandem WH2 motifs close to actin in raw coordinates, but the model does not support those contacts with chain-pair confidence. The high tandem x5 overall ipTM is therefore dominated by actin-actin assembly, not trusted WH2-actin placement.
  • Limitations: AF3 is weak for short-motif actin side binding. Phase 4k should be treated as a negative-control/geometry caution, not as a falsification of the Phase 4m Kd soft-floor result.

Ranking delta

  • Hypothesis h09: tier S S | mech 5 5 | deliv 4 4 | misha_fit 3 3
  • Evidence: Phase 4k does not independently close the N>=4 geometry gate, but Phase 4m already closed the higher-priority Kd_mono gate.
  • Next step changes to: Phase 4l remains optional unless a paper-grade C_eff estimate is needed.

Connections