Pharmacochaperone (h01) — parameter provenance
Source agent: h01 audit (Sonnet 4.6), 2026-04-23. Consumer: all 17 pharmacochaperone_*.py scripts in models/.
Parameter table
| Parameter | Value used | Units | Source | Page/section | Status | Script(s) | Action |
|---|---|---|---|---|---|---|---|
| Water probe radius | 1.4 | Å | Physical constant (Lee & Richards 1971) | standard | ✅ physical | phase1_mutant_pocket | None |
| Carbon vdW radius | 1.7 | Å | Physical constant (Bondi 1964) | standard | ✅ physical | phase1_mutant_pocket | None |
| Salt-bridge O-N distance | 2.8 | Å | Physical constant (Hubbard geometry) | standard | ✅ physical | phase3c_shape_fit | None |
| K-to-kJ conversion | 1/4.184 | — | Physical constant | — | ✅ physical | phase4f | None |
| Water dielectric constant | 78.5 | — | Physical constant | — | ✅ physical | phase4f (eps_solvent) | None |
| pLDDT threshold | 70 | — | Standard AF2/AF3 convention (Jumper 2021 Nature) | — | ✅ standard | phase1b | None |
| AmberFF19SB | — | — | He 2020 JCTC | — | ✅ standard MD | phase5 | None |
| GAFF2 | — | — | Wang 2004 JCCA | — | ✅ standard MD | phase5 | None |
| TIP3P water | — | — | Jorgensen 1983 JCP | — | ✅ standard MD | phase5 | None |
| NaCl concentration | 0.15 | M | Physiological | — | ✅ standard | phase5 | None |
| Rule-of-3 (Congreve): MW<300, logP<3, HBD≤3, HBA≤3, rot≤3 | — | — | Congreve 2003 Drug Discov Today | — | ✅ standard (textbook) | phase3b | Optional citation note |
| Docking box size | 18×18×18 | Å | In-silico: Phase 2B subpocket volume → box calibrated to 159 ų pocket | — | ✅ in-silico | target_prep, phase4_common | None |
| K1141-NZ to F1646-ring distance | 5.69 → 5.7 | Å | In-silico: measured from AF3 WT CIF | Phase 3A | ✅ in-silico | phase3b | None |
| Reference ΔΔG gap | 8.4 | kcal/mol | In-silico: STRC Electrostatic Analysis E1659A (multi-method PAE-corrected) | own note | ✅ in-silico | phase4f (REFERENCE_GAP_KCAL_MOL) | Add # see STRC Electrostatic Analysis E1659A comment |
| Druggability threshold (Phase 4a gate) | ≥0.70 | 0-1 | In-silico: calibrated to Phase 0 result (0.86 WT) | Phase 0 | ✅ in-silico | phase4a | None |
| ΔΔG gate (WT vs MUT docking) | 1.0 | kcal/mol | In-silico: ~2× Vina paired noise (0.5-0.8 kcal/mol) | phase4c comment | ✅ in-silico | phase4c | None |
| K1141A score-loss gate | ≥2.0 | kcal/mol | In-silico: 2× DELTA_GATE, conservative pharmacophore falsification | — | ✅ in-silico | phase4d | None |
| Vina exhaustiveness | 32 | — | Standard Vina: 4× default (8), conservative for small pocket | Trott 2010 JCIM; Eberhardt 2021 JCIM (software version) | ✅ standard | phase4b | None |
| Vina score threshold (diflunisal gate) | ≤−5.0 | kcal/mol | Pipeline-specific positive-control gate: diflunisal must score ≤ −5 to validate the pocket. NOT a CASF universal threshold. Eberhardt 2021 defines no kcal/mol cutoff. | Eberhardt 2021 (software ref); threshold is in-silico heuristic | ✅ labeled in-silico — gate documented [audit 2026-04-25] | phase4b | Docstring updated: cite Eberhardt 2021 as software version; threshold labeled pipeline-specific |
| MM-PBSA ΔG_bind gate | ≤−6.0 | kcal/mol | Pipeline-specific empirical gate providing ~2σ separation from non-binder baseline (Genheden 2015: std error 2.6–3.3 kcal/mol). NOT a universal Genheden-derived cutoff; no such universal cutoff exists. | Genheden & Ryde 2015 Expert Opin Drug Discov (PMC4487606) — characterizes error bands | ✅ labeled pipeline-specific with error-band justification [audit 2026-04-25] | phase5 | Docstring updated: cite Genheden 2015 for error bands; gate labeled pipeline-specific |
| RWM PK envelope: MW 250-400 Da, LogP 1.5-3.5, HBD 0-2 | per range | — | 2018-salt-plontke-pharmacokinetic-principles-inner-ear | Discussed qualitatively | ✅ paper in strc/papers/ | target_prep docstring | None |
| TPSA — descriptor only, no filter applied | n/a | Ų | TPSA computed and stored as descriptor in phase3b output. NOT used in composite score. “CNS-like 40-90” framing was incorrect (STRC extracellular; BBB irrelevant). Hypothesis note 70-100 Ų cites Salt & Plontke 2018 qualitatively; no primary otic/RWM TPSA range exists in primary literature. | Salt & Plontke 2018 (qualitative); no primary otic TPSA range | ✅ descriptor-only; no filter range applied [fixed audit 2026-04-23; confirmed audit 2026-04-25] | phase3b | None — docstring already corrected 2026-04-23 |
| MW filter for virtual screen (score_size) | 180–350 | Da | Fragment-library pocket-fit scoring function. 350 Da = upper fragment-drug boundary matching 159 ų pocket. Distinct from RWM delivery envelope (200-500 Da, hypothesis note). Two uses; labeled separately. | In-silico: pocket volume 159 ų → fragment MW ceiling; hypothesis note RWM range is delivery, not pocket-fit | ✅ labeled distinct contexts [audit 2026-04-25] | phase3b | Docstring updated: note score_size is pocket-fit (180-350 Da), delivery PK envelope is 200-500 Da per hypothesis note |
| Druggability v_opt | 250 ų (phase2b subpockets), 300 ų (phase2 full pocket) | ų | Halgren 2009 SiteMap abstract confirms composite druggability scoring concept; full issue PDF downloaded via Anna’s Archive (MinerU parsing). v_opt difference physically justified: phase2 scores full pocket clusters, phase2b scores subpockets. Both labeled SiteMap-inspired heuristic. | Halgren 2009 JCIM (PMID 19434839) — conceptual source; full text in parsing | ✅ SiteMap-inspired heuristic; v_opt difference justified by scale [audit 2026-04-25] | phase2 (300), phase2b (250) | Docstrings updated with Halgren 2009 PMID and “in-house approximation” label [2026-04-25] |
| Druggability composite weights | 0.5/0.3/0.2 (p1), 0.4/0.25/0.2/0.15 (p2), 0.3/0.2/0.15/0.15/0.2 (p2b) | — | Three schemes reflect different feature sets per phase: p1 has 3 features; p2 adds nres; p2b adds depth. Halgren 2009 formula inaccessible. Each scheme is within-phase only; cross-phase comparison is invalid. | Halgren 2009 (conceptual); weights in-house | ✅ three-scheme inconsistency documented; cross-phase incomparability declared [audit 2026-04-23; confirmed 2026-04-25] | phase1, phase2, phase2b | None beyond existing docstring flags |
| MM-GBSA recovery gate | ≥30% of 8.4 kcal/mol | — | Clinical analogy: VX-809 restores ~30% CFTR function | Bulawa 2012 PNAS | ✅ clinical analogy — defensible | phase4f (RECOVERY_GATE) | None; note already references Bulawa class implicitly |
Resolved flags [audit 2026-04-25]
All 6 ⚠ rows above closed. Summary:
- TPSA — descriptor-only (no filter). “CNS 40-90” framing removed 2026-04-23. No primary otic range exists; no fabricated range added. ✅
- Druggability weights — three-scheme inconsistency documented in all three script docstrings 2026-04-23. Each phase uses within-phase scoring only; cross-phase comparison flagged as invalid. ✅
- Vina −5 kcal/mol gate — relabeled as pipeline-specific positive-control gate; Eberhardt 2021 cited as software-version reference. Gate does not claim to derive from a universal CASF threshold. ✅
- MM-PBSA −6 kcal/mol gate — relabeled as pipeline-specific empirical gate; Genheden & Ryde 2015 cited for error-band context (~2.6–3.3 kcal/mol std error). ✅
- MW filter 180-350 vs 200-500 — labeled as distinct contexts (score_size = pocket-fit fragment scoring; 200-500 = RWM delivery envelope in hypothesis note). Both valid; documented separately. ✅
- Druggability v_opt 250 vs 300 — v_opt difference physically justified by scale (subpocket vs full-pocket scoring). Halgren 2009 cited as conceptual source; full formula paywalled. ✅
No phantom citations found
Unlike h09, h01 scripts contain zero fabricated paper citations. No “Salt 2011” style phantom references. Every numeric constant traces to either a physical constant, an in-silico output, a standard community convention, or a real paper already in strc/papers/. The issues above are internal inconsistencies and missing attribution notes — not citation fabrications.
Methods textbook references (added 2026-04-25)
For citation provenance on de novo design / FBDD / scoring-function classification / efficiency metrics, the canonical methods textbook is now ingested:
| Reference | Role | Local note |
|---|---|---|
| Schneider (Ed.) 2014, De novo Molecular Design, Wiley-VCH | de novo / FBDD / scoring-function classes / efficiency metrics / FE-method choice | 2014-schneider-de-novo-molecular-design-book |
Used to back the following h01 docstring claims (citation pattern: “Schneider 2014 §X.Y”):
- Rule-of-three filter (phase 3b) — §5.2.2 / Table 5.1
- Vina = force-field-class scoring (phase 4b) — §1.6.1 Eq. 1.10
- MM-GBSA = force-field + continuum-correction (phase 4f) — §1.6.1 + §16.2.5.2
- LE-based fragment prioritization — §6.4.1 / Eq. 1.8
- Fragment-grow strategy default for h01 v4 — §5.3.4 Fig. 5.1c + §6.5
- Ligandability vs druggability discipline — §6.3.10 + Fig. 6.5
Recipes available (P1):
- Recipe — Receptor-Based Scoring Function Selection
- Recipe — Fragment Library Filtering Pipeline
- Recipe — Fragment Optimization Linking Merging Growing
Reference data (P0):
- Lipinski Rule of Fives vs Congreve Rule of Threes Reference Table — verbatim Table 5.1
- De Novo Design Software Scoring Strategy Catalog — verbatim Table 1.3
- Ligand Efficiency Metrics Catalog — verbatim §6.4.1 formulas
Concepts (P2):
- Fragment Additivity Assumption and Superadditivity — non-additivity warning for fragment-grow
- Druggability vs Ligandability Distinction — sharpens what h01 phase 2 / 2b actually measures
Connections
[part-of]_hub (literature-params)[see-also]2018-salt-plontke-pharmacokinetic-principles-inner-ear — RWM PK envelope (needed for TPSA fix)[see-also]STRC Electrostatic Analysis E1659A — source of 8.4 kcal/mol reference gap[see-also]STRC Pharmacochaperone Virtual Screen E1659A — main h01 hypothesis note[see-also]2014-schneider-de-novo-molecular-design-book[applies]STRC Hypothesis Ranking (h01, A-tier)