STRC Research

STRC h01 Phase 9x v5.3 Lead Committee Alpha Sensitivity 2026-04-27

2 min read

STRC h01 Phase 9x — v5.3 Lead Committee Alpha Sensitivity

v5.3 rank aggregation is alpha-robust: adamantyl_SO2Me_Cl stays #1 for alpha=0.5-2.0 across Boltz, Vina, tauRAMD, and ADMET axes.

Problem

The v5.3 committee has a method-class split: 1-indanyl_acylsulfonamide_SO2Me_-Cl is the affinity-led candidate, while adamantyl variants look cleaner on ADMET and kinetic-confidence axes. This proof asks whether the committee decision is fragile to the kinetic-axis weight alpha, using only already-delivered Phase 5q / Phase 5e / Phase 8c data.

Method

  • Script: hypotheses/h01-pharmacochaperone/scripts/phase9x_v53_alpha_sensitivity.py
  • Inputs:
    • Boltz-2: models/boltz_jobs_2026-04-26_phase5q_v53/aggregate.json
    • Mutant Vina ensemble: hypotheses/h01-pharmacochaperone/artifacts/phase5e_v2/v5_3-top3/aggregate.json
    • STRC tauRAMD n=20: hypotheses/h01-pharmacochaperone/artifacts/phase5q_strc_v53/ranking.json
    • ADMET-AI v5.3: models/artifacts/phase8c_v5_3_admet/v5_3_admet_summary.json
  • Rank model: score = mean_affinity_rank + alpha * mean_kinetic_rank + mean_admet_rank. Lower score is better.
  • Alpha grid: 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0.

Axes

BlockRank axes
AffinityBoltz rank0 ipTM higher, Boltz 5-seed mean ipTM higher, Vina mean dG lower, Vina median dG lower, Vina best dG lower
KinetictauRAMD mean higher, tauRAMD median higher, tauRAMD SEM lower
ADMETgate flags lower, max endpoint percentile lower, mean endpoint percentile lower

All axes are ranks, not physical-unit sums; this avoids pretending kcal/mol, ipTM, ps, and ADMET percentiles are commensurate.

Results

alpharank 1scorerank 2scorerank 3score
0.50adamantyl_acylsulfonamide_SO2Me_-Cl4.100adamantyl_acylsulfonamide_SO2Me_-CF34.9671-indanyl_acylsulfonamide_SO2Me_-Cl5.933
0.75adamantyl_acylsulfonamide_SO2Me_-Cl4.517adamantyl_acylsulfonamide_SO2Me_-CF35.3831-indanyl_acylsulfonamide_SO2Me_-Cl6.600
1.00adamantyl_acylsulfonamide_SO2Me_-Cl4.933adamantyl_acylsulfonamide_SO2Me_-CF35.8001-indanyl_acylsulfonamide_SO2Me_-Cl7.267
1.25adamantyl_acylsulfonamide_SO2Me_-Cl5.350adamantyl_acylsulfonamide_SO2Me_-CF36.2171-indanyl_acylsulfonamide_SO2Me_-Cl7.933
1.50adamantyl_acylsulfonamide_SO2Me_-Cl5.767adamantyl_acylsulfonamide_SO2Me_-CF36.6331-indanyl_acylsulfonamide_SO2Me_-Cl8.600
1.75adamantyl_acylsulfonamide_SO2Me_-Cl6.183adamantyl_acylsulfonamide_SO2Me_-CF37.0501-indanyl_acylsulfonamide_SO2Me_-Cl9.267
2.00adamantyl_acylsulfonamide_SO2Me_-Cl6.600adamantyl_acylsulfonamide_SO2Me_-CF37.4671-indanyl_acylsulfonamide_SO2Me_-Cl9.933

Artifacts:

  • hypotheses/h01-pharmacochaperone/artifacts/phase9x_v53_alpha_sensitivity/alpha_sensitivity.json
  • hypotheses/h01-pharmacochaperone/artifacts/phase9x_v53_alpha_sensitivity/input_metrics.csv
  • hypotheses/h01-pharmacochaperone/artifacts/phase9x_v53_alpha_sensitivity/alpha_scores.csv
  • hypotheses/h01-pharmacochaperone/artifacts/phase9x_v53_alpha_sensitivity/alpha_sensitivity.md

Verdict

  • Outcome: SUPPORTIVE.
  • Key number: adamantyl_acylsulfonamide_SO2Me_-Cl remains rank 1 for all alpha values from 0.5 to 2.0.
  • Interpretation: The practical v5.3 lead committee is not fragile to kinetic-axis weighting. The affinity-led 1-indanyl candidate is useful as a method-class anchor, but it loses once ADMET and kinetic-confidence are included by rank.
  • Limitation: This is a committee robustness check, not a new physical binding-free-energy calculation. It reuses existing Phase 5q / Phase 5e / Phase 8c outputs.

Ranking delta

  • Hypothesis h01: tier A → A | mech 4 → 4 | deliv 3 → 3 | misha_fit 4 → 4.
  • No hub frontmatter change. This supports the current “no single winner / adamantyl-Cl practical lead” interpretation but does not change the mechanism or delivery score.

Connections

Graph View

Backlinks