STRC h01 Phase 8c v5.3 ADMET-AI Triage 2026-04-26
ADMET-AI (Swanson 2024) percentile-rank predictions on the 3 v5.3 acyl-sulfonamide PASS-list ligands from STRC h01 Phase 5p v53 Acyl Sulfonamide Library 2026-04-26 / STRC h01 Phase 5q v5.3 Acyl Sulfonamide Boltz-2 + Vina Consensus 2026-04-26. Same 10-endpoint gate at the 90th drugbank-approved percentile as Phase 8c (v5) and Phase 8f (v5.2).
Method
- Tool: ADMET-AI 1.4.x via
/opt/miniconda3/envs/admet-ai/bin/admet_predict. Apple Silicon MPS available but ADMET-AI uses CPU forward pass for the Lightning ensembles (sub-second per molecule × 41 endpoints). - Input: 3 SMILES from
~/STRC/hypotheses/h01-pharmacochaperone/artifacts/phase5p/v53_library.csv, neutral form (Gasteiger / RDKit canonical, anion not deprotonated for ADMET-AI which expects neutral SMILES). - Gate definition (matches Phase 8c/8f):
*_drugbank_approved_percentile ≥ 90.0onhERG, BBB_Martins, DILI, AMES, Carcinogens_Lagunin, ClinTox, Bioavailability_Ma, CYP3A4_Veith, Clearance_Hepatocyte_AZ, LD50_Zhu⇒ flag.
Results
Per-ligand summary
| Ligand | flags | hERG | BBB | DILI | AMES | Carc | ClinTox | Bioavail | CYP3A4 | ClrHep | LD50 |
|---|---|---|---|---|---|---|---|---|---|---|---|
1-indanyl_acylsulfonamide_SO2Me_-Cl (Boltz/Vina LEAD) | 1 | 65.2 | 53.7 | 85.7 | 49.2 | 73.0 | 52.3 | 45.3 | 90.4* | 12.7 | 80.2 |
adamantyl_acylsulfonamide_SO2Me_-Cl (τ tightest SEM) | 0 | 66.0 | 55.5 | 84.0 | 3.2 | 60.2 | 34.4 | 49.4 | 87.0 | 23.7 | 69.3 |
adamantyl_acylsulfonamide_SO2Me_-CF3 (τ rank-1) | 0 | 68.8 | 58.1 | 82.0 | 2.0 | 51.6 | 34.0 | 50.1 | 87.1 | 12.1 | 85.8 |
* = above 90 percentile gate.
Class observations
- hERG cluster 65–69 percentile. All three ligands sit in the same hERG band, well below the 90 gate. No liability separation between scaffolds on this axis. Consistent with the Phase 8a v5 design rationale — sulfonamide head reduces basic-amine driven hERG affinity vs the v3b-class.
- AMES separation. 1-indanyl_-Cl 49.2 vs adamantyl scaffolds 2–3 percentile. Aromatic indane ring is a known AMES-positive scaffold heuristic, predicted shift is consistent.
- CYP3A4 90.4 on 1-indanyl_-Cl. Marginal; rounds to gate. Adamantyl scaffolds 87.0 / 87.1 — within 3 percentile units. Class-wide CYP3A4 induction risk is real but borderline; this single flag would not disqualify the molecule on its own.
- DILI band 82–86, ClinTox 34–52, BBB 53–58. All three sit in the same hepatic / clinical-tox / brain-penetration window. No scaffold separates the trio on these axes.
Lead-committee tradeoff (paper-grade)
| Axis | rank-1 | rank-2 | rank-3 |
|---|---|---|---|
| Phase 5q Boltz-2 ipTM (affinity, dynamic) | 1-indanyl_-Cl 0.645 | adamantyl_-Cl 0.589 | adamantyl_-CF3 0.578 |
| Phase 5q Vina-v3 best ΔG (affinity, rigid) | 1-indanyl_-Cl −4.59 | adamantyl_-Cl −5.60 | adamantyl_-CF3 −4.60 |
| Phase 5e_v2 mut-ensemble mean ΔG (n=20) | 1-indanyl_-Cl −0.67 | adamantyl_-Cl +4.09 | adamantyl_-CF3 +4.54 |
| Phase 5q τRAMD mean (residence, biased MD) | adamantyl_-CF3 23.88 | 1-indanyl_-Cl 18.00 | adamantyl_-Cl 16.46 |
| Phase 5q τRAMD SEM (replicate confidence) | adamantyl_-Cl 0.65 | 1-indanyl_-Cl 2.76 | adamantyl_-CF3 5.23 |
| Phase 8c v5.3 ADMET clean (0 flags) | adamantyl_-Cl, adamantyl_-CF3 | (tied) | 1-indanyl_-Cl (1 flag) |
No single ligand wins all axes. The conservative lead committee is two-track:
- Affinity-led track:
1-indanyl_acylsulfonamide_SO2Me_-Cl. Strongest mut-ensemble docking (mean −0.67 vs adamantyl mean +4.09/+4.54), Boltz-2 + Vina rank-1 by both methods. ADMET cost = 1 borderline CYP3A4 flag. - ADMET/kinetic-confident track:
adamantyl_acylsulfonamide_SO2Me_-Cl. Fully ADMET-clean, lowest τRAMD SEM (0.65 ps replicate confidence), but mut-ensemble mean ΔG +4.09 means it is NOT a robust pocket binder — relies on best-pose selection.
For the paper, both tracks should be reported. 1-indanyl_-Cl is the affinity lead; adamantyl_-Cl is the ADMET-and-kinetic-confidence backup. The empirical kinetic-selectivity verdict (Phase 5m + Phase 5q) applies to the entire class regardless of within-class ranking.
Ranking delta
A held. mech 4 / deliv 3 / misha_fit 4 unchanged. ADMET axis closes the lead-committee tradeoff structurally: there is no v5.3 ligand that is simultaneously affinity-rank-1 + τ-rank-1 + ADMET-clean. This is a scaffold-class limit, not a within-class ranking failure. v5.4 design (next major version) would need to break the affinity↔ADMET tradeoff at the head-group / linker level.
- tier A → A
- mech 4 → 4
- deliv 3 → 3
- misha_fit 4 → 4
- next_step refined: lead committee = (affinity-track 1-indanyl_-Cl, ADMET-track adamantyl_-Cl). Both belong on the paper Figure 5 lead summary. v5.4 design rationale = break affinity↔ADMET tradeoff at SO2Me / sulfonamide-N linker (not chlorine substituent).
Connections
[part-of]h01 hub[applies]STRC h01 Phase 5q v5.3 Acyl Sulfonamide Boltz-2 + Vina Consensus 2026-04-26[applies]STRC h01 Phase 5q v5.3 STRC Within-Target tauRAMD 2026-04-26[applies]STRC h01 Phase 5p v53 Acyl Sulfonamide Library 2026-04-26[contrasts]STRC h01 Phase 8c v5 ADMET-AI Triage 2026-04-24[see-also]STRC Computational Scripts Inventory