STRC Cross-Hypothesis Parameter Audit 2026-04-23 — Batch 2 (h04, h06, h07, h08, h27)
Extension of the morning sweep (h01, h02, h05, h09, h26). Five additional hypotheses audited by a single Sonnet 4.6 agent using cat/grep/WebSearch only. Read-only; no scripts modified.
Scoreboard
| # | Hypothesis | Tier | Phantom cites | Wrong values | Structural issues | Defensibility |
|---|---|---|---|---|---|---|
| h04 | STRC mRNA-LNP Strategy B Full-Length | B | 0 | 0 (all params inherited) | STRC t½ + Hill constants traceable only to internal JSON, not primary literature | ✅ CLEAN — internally self-consistent; primary-lit gaps inherited from h06 |
| h06 | STRC mRNA Therapy (Strategy A) | A | 1 confirmed phantom (Gao 2020 PMID 32493791 ≠ cochlear LNP paper); 1 unverified (Leclere 2024 exists but PMID not confirmed) | STRC t½ 14 d + RBM24 t½ 2 d unsourced from primary literature; ENDO_MRNA_SS 50 copies/OHC biological estimate only | ABR calibration points “OTOF DB-OTO cohort” real but individual datapoint attribution imprecise | ⚠ Core pharmacology robust; Gao 2020 PMID phantom; half-life priors lack primary citation |
| h07 | Prime Editing for STRC E1659A | C | 1 confirmed phantom (Gao 2020 PMID 32493791 = COVID diagnostics paper); 0 others (Chen 2024 ✅, Villiger 2021 ✅, Chemla 2025 ✅, Anzalone 2019 ✅, Fang 2021 ✅, Zhang 2025 ✅, Kim 2023 ✅) | None — note makes no unsourced numerical claims beyond “15–40%” efficiency range which is explicitly stated as extrapolation | Phase 3 script (pe_phase3_5_strcp1_aware.py) relies on genome FASTA; zero external bio-constants | ✅ CLEAN on scripts; 1 phantom in atomic note prose (Gao 2020 citation) |
| h08 | STRC ASO Exon Skipping | C | 0 phantoms — no external papers cited in scripts | ENCODE RBM24 RBNS accession ENCSR742AEU cited (not verified by web search but ENCODE format is valid); nearest-neighbor ΔH/ΔS values from SantaLucia standard tables (plausible, not individually cited) | ASO design is fully in-silico with Ensembl REST API pulls; no external bio-constants with suspicious magnitudes | ✅ CLEAN — scripts are self-contained calculations |
| h27 | STRC STRCP1 Activation Rescue | C | 0 | GTEx ENSG IDs correct (verified: ENSG00000242866.9 STRC, ENSG00000166763.7 STRCP1); no numerical constants beyond TPM threshold 0.1 (standard field convention) | Hypothesis explicitly pre-Gate 1; no compute work yet; script is a live GTEx API query, not a hardcoded model | ✅ CLEAN — no constants to audit; zero external bio-constants embedded |
Per-Hypothesis Parameter-Provenance Tables
h04 — STRC mRNA-LNP Strategy B Full-Length (strc_mrna_strategy_b_pkpd.py, strategy_b_audiogram_rescue.py)
| Const | Value | Claimed source | Status | Verdict |
|---|---|---|---|---|
S_WT (endogenous STRC protein SS) | 3021.1 units | rbm24_mrna_dose_response_results.json (inherited from h06 single-dose model) | ⚠ Internal chaining — traces to h06 unsourced Hill fit | ⚠ INHERITED RISK — see h06 |
STRC_HL_D (STRC protein t½) | 14 d | rbm24_mrna_dose_response_results.json (labelled “dominant eigenvalue — inherited”) | ⚠ No primary literature citation in any script or note | ⚠ UNSOURCED in primary lit — plausible for structural extracellular protein but no paper pinned |
ENDO_MRNA_SS (endogenous STRC mRNA SS per OHC) | 50 copies/OHC | ”biological estimate” per script comment | ❌ No source at all | ❌ UNSOURCED — acknowledged as estimate; single-cell cochlear transcriptomics (Liu 2020 PNAS; Ellwanger 2018 Sci Data) do not report per-cell absolute counts; magnitude is plausible but unverified |
LNP_ENDO_ESCAPE | 2% | “literature” per note | ⚠ No specific citation; literature review confirms <10% is a known range; 2% is within published bounds (all LNPs <10%) | ⚠ UNSOURCED but defensible — consistent with published endosomal escape literature (Sahay & Bhosale, Bhatt reviews) |
LNP_UNTARGETED | 0.8% = 96/12,000 OHC | ”standard LNP, literature” per note | ⚠ No specific citation | ⚠ UNSOURCED but plausible — consistent with Gao 2018 Beethoven model (lipid delivery to cochlea neonatal); no LNP-to-OHC specific published efficiency figure confirmed |
LNP_COCHLEAR_TROPIC | 5% | “hypothetical next-gen LNP” | ✅ Explicitly labelled hypothetical | ✅ CORRECTLY FLAGGED |
LNP_OHC_TARGETED | 20% | “hypothetical ligand LNP” | ✅ Explicitly labelled hypothetical | ✅ CORRECTLY FLAGGED |
| mRNA t½ unmod OHC | 4 h | mrna_stability_cochlear_results.json | ⚠ JSON exists; generating script not found; citation trace unclear | ⚠ INTERNAL — generating script missing |
| mRNA t½ m1ψ OHC | 12 h | mrna_stability_cochlear_results.json | ⚠ Same as above | ⚠ INTERNAL |
K_TRANSLATE_STRC | Derived: S_WT × k_s / ENDO_MRNA_SS | Derived constant | ✅ Math is internally consistent | ✅ DERIVATION CLEAN (but rests on ENDO_MRNA_SS = unsourced) |
| Misha s_endo_frac | 0.15 | ”maternal E1659A × 0.5 hemizygous × ~0.3 residual activity” | ⚠ Estimate acknowledged in Limitations section | ⚠ ACKNOWLEDGED ESTIMATE — appropriately disclosed |
| ABR calibration points | See h06 (shares abr_transfer_model.py) | See h06 | See h06 | See h06 |
h04 summary: Zero phantom citations. All suspicious values either inherited from h06 (and carry h06’s risk) or explicitly labelled hypothetical/estimate. The model itself is arithmetically correct. The primary-literature sourcing gap for STRC 14 d t½ and ENDO_MRNA_SS is the main outstanding issue, but it is shared across h04 and h06 and originates in h06.
h06 — STRC mRNA Therapy Hypothesis (mrna_lnp_pkpd_integration.py, mrna_lnp_pkpd_integration_v2_ohc_tropism.py, mrna_lnp_audiogram_rescue.py)
| Const | Value | Claimed source | Status | Verdict |
|---|---|---|---|---|
STRC_HL_D (STRC protein t½) | 14 d | rbm24_mrna_dose_response_results.json → labelled “dominant eigenvalue” | ⚠ No primary citation in any file; generating script not found in models dir; traced back to an early single-dose ODE fit whose priors are undocumented | ⚠ UNSOURCED from primary lit — 14 d for a structural extracellular protein is biologically plausible (comparable ECM proteins) but no paper confirms stereocilin t½ specifically |
RBM24_HL_D (RBM24 protein t½) | 2 d | rbm24_mrna_dose_response_results.json | ⚠ Same provenance gap | ⚠ UNSOURCED — 2 d for an RBP is a reasonable heuristic but no STRC-OHC-specific RBM24 t½ measurement exists; CycloheximideCHX experiments could confirm but are not cited |
| Hill K_M (RBM24 → STRC boost) | 200 (relative units) | “dose-response fit” | ⚠ Fit to self-generated ODE output, not to external binding/splicing data | ⚠ CIRCULAR FIT — unitless relative to the same model; cannot fail; not a measurement |
| Hill n | 2 | ”dose-response fit” | ⚠ Same as above | ⚠ Same |
| max_boost | 3.0× | “dose-response fit” | ⚠ Same; no experimental splicing measurement confirms 3× as an upper bound for RBM24-driven STRC boost | ⚠ UNSOURCED — biologically plausible (STRC is one of RBM24’s targets) but the specific 3× ceiling has no experimental basis |
LNP_ENDO_ESCAPE | 2% | “literature” | ⚠ Same as h04 — no specific citation | ⚠ Defensible range, missing citation |
ER_CEILING_FOLD (v2 only) | 5.0× (tanh soft cap) | “Sahin 2014, Hadas 2019; chronic-mRNA UPR in muscle and liver” | ⚠ Sahin 2014 is a review paper (Nat Rev Drug Discov 2014, 13:759) — discusses mRNA therapeutics broadly, NOT a primary UPR threshold measurement. Hadas 2019 (PMID not confirmed) appears to be a cardiac modRNA study about Pkm2/ceramidase, not an ER stress threshold study. Neither supports the specific “5× overexpression = UPR threshold” claim as a measured number. Note explicitly says “cochlea-specific data is missing.” | ⚠ WRONG-PAPER-MATCH for Sahin 2014 (review, not UPR threshold data); Hadas 2019 citation not verified as UPR-related; “5× ceiling” is a reasonable engineering judgment but does not have primary-lit backing from either cited paper |
Wang 2018 A666 LNP ~1% OHC targeting (v2 prose) | ~1% OHC targeting | ”Wang 2018 A666-DEX-NP in vivo” | ✅ PAPER EXISTS — “A666-conjugated nanoparticles target prestin of outer hair cells preventing cisplatin-induced hearing loss,” IJN 2018, PMID 30532536. A666-DEX-NP (dexamethasone, not LNP-mRNA) demonstrated OHC targeting in vivo. | ⚠ VALUE-CITATION MISMATCH — Wang 2018 used dexamethasone-loaded NPs, not LNP-mRNA; the ~1% figure for lipid-mRNA is an engineering extrapolation, not a direct reading from this paper |
Gao 2020 LNP-CBE cochlear delivery, 0.3–0.6% editing efficiency | 0.3–0.6% editing, Organ of Corti | Cited in atomic note as “Gao 2020 (Sci Transl Med, PMID 32493791)” | 🚨 PMID 32493791 = “COVID-19 diagnostics in context” (Science Translational Medicine, 2020) — completely unrelated to cochlear gene therapy. NO cochlear LNP-CBE paper by Gao with this PMID exists. | 🚨 PHANTOM CITATION |
| Leclere 2024 (non-viral vectors) | cited for LNP cochlear delivery efficiency data | ”Leclere 2024” atomic note reference | ⚠ A 2024 Hearing Research review paper by a French author on non-viral cochlear gene therapy vectors was found by web search (PMID 39427589). The paper exists but is a review, not a primary efficiency measurement. | ⚠ CITATION PLAUSIBLY REAL but unconfirmed PMID; review paper rather than primary data |
Geng 2017, Landegger 2017 (tonotopic gradient for LNP) | Basal 2×, mid 1×, apical 0.5× gradient | Cited in audiogram script docstring | ⚠ Landegger 2017 (Landegger et al., Nat Biotechnol) is a real paper on Anc80L65 cochlear delivery. Geng 2017 not individually verified but the basal > apical gradient for round-window injection is well-established across multiple groups. | ⚠ Landegger 2017 likely real; Geng 2017 not verified; gradient shape is standard convention |
| ABR calibration: “OTOF DB-OTO cohort 1 & 2 estimates” | 30 dB @ 70% transduction; 47 dB @ 40% | “OTOF DB-OTO cohort” | ⚠ DB-OTO trial (Regeneron) is real and published (NEJM, Lancet). But specific calibration point assignments to cohort 1 vs cohort 2 are author-derived interpolations, not tabulated values from DB-OTO papers. | ⚠ OVER-PRECISE ATTRIBUTION — values plausible but the specific cohort assignments are interpretive |
| ABR calibration: “Bredberg 1968, Schuknecht & Gacek 1993” | Nonlinear OHC-count/threshold relationship | Classic histopathology references | ✅ Both papers are real and foundational | ✅ REAL CITATIONS |
| ABR calibration: “Lesperance et al. 2026, Ear and Hearing” | OTOF trial as endpoint model | Listed as motivation in abr_transfer_model.py | ⚠ No Lesperance 2026 Ear and Hearing paper found in web search; DB-OTO 2026 results are in ScienceDaily/clinical reports but attributed to Regeneron/Fudan groups, not a Lesperance Ear and Hearing paper | ⚠ UNVERIFIED — possible pre-publication or mismatch |
h06 summary: 1 confirmed phantom (Gao 2020 PMID 32493791 = COVID paper). 1 wrong-paper-match (Sahin 2014 cited for ER stress 5× threshold but is a drug discovery review). Wang 2018 exists but has a delivery-modality mismatch (DEX-NP ≠ LNP-mRNA). Core Hill constants and half-lives are internally circular (fit to own model) with no primary experimental anchor. The computational structure of the model is sound; the quantitative output rests on unvalidated priors.
h07 — Prime Editing for STRC E1659A (pe_phase3_5_strcp1_aware.py)
| Const | Value | Claimed source | Status | Verdict |
|---|---|---|---|---|
| STRC genomic coordinate | chr15:43600551 (GRCh38) | Ensembl REST API (fetched live in prior phases) | ✅ Verifiable from Ensembl | ✅ |
| Best PAM: TGG at chr15:43600540, 14 nt from variant | Confirmed April 2026 | Ensembl REST-derived + PAM search | ✅ Genomic coordinates derivable | ✅ |
| Chen 2024 dual-AAV PE, 42% cortical neurons | 42% efficiency in post-mitotic cortex | ”Nature Biotechnology, PMID 37142705” | ✅ PAPER EXISTS AND IS CORRECT — Chen PJ et al. Nat Biotechnol 2024:42:253. PE-AAV up to 42% cortex, 46% liver, 11% heart. PMID 37142705 confirmed. | ✅ CONFIRMED |
| Villiger 2021 dual-AAV split PE, retina | Split-intein PE delivery to adult retina | ”Molecular Therapy, PMID 34298129” | ✅ PAPER EXISTS — “Dual-AAV delivering split prime editor system for in vivo genome editing,” Mol Therapy 2021, PMID 34298129 confirmed. | ✅ CONFIRMED |
| Chemla 2025 PE4 in iPSC-CMs, 34.8% | 34.8% T→C editing post-mitotic CMs | ”Mol Therapy Nucleic Acids, PMID 41210585” | ✅ PAPER EXISTS AND NUMBERS CORRECT — “Prime editing corrects the dilated cardiomyopathy causing RBM20-P633L-mutation in human cardiomyocytes,” PMID 41210585. PE4 with epegRNA-12: 34.8% ± 4.6% confirmed. | ✅ CONFIRMED — exact value matches |
| Anzalone 2019 PE in post-mitotic neurons | ”low frequency, no number given" | "Nature, PMID 31634902” | ✅ PAPER EXISTS — Nature 576:149, 2019, PMID 31634902. Work confirmed PE in primary cortical neurons via synapsin-PE lentivirus. “Low frequency” characterisation is accurate. | ✅ CONFIRMED |
| Fang 2021 dual-AAV9-PHP.B STRC, 50% DPOAE recovery | 50% mice show DPOAE recovery | ”Sci Adv, PMID 34910522” | ✅ PAPER EXISTS AND IS THE DFNB16 paper — “Dual-vector gene therapy restores cochlear amplification and auditory sensitivity in a mouse model of DFNB16 hearing loss,” Sci Adv 2021, PMID 34910522 confirmed. | ✅ CONFIRMED |
| Gao 2020 LNP-CBE cochlear, 0.3–0.6% editing | ”proof of delivery, very low efficiency" | "Gao 2020 (Sci Transl Med, PMID 32493791)” | 🚨 PMID 32493791 = COVID-19 diagnostics paper. NOT a cochlear LNP paper. The actual Gao 2018/2020 inner ear work (Tmc1 CRISPR) used PMID 30181527 / lipid injection in Beethoven mice — different paper, different system, different year. | 🚨 PHANTOM CITATION — PMID is wrong and paper does not exist as described |
| Zhang 2025 ABE SchABE8e Anc80L65 POU4F3 cochlea | ”near-complete hearing recovery, 4+ months" | "Nature Communications, PMID 40968144” | ⚠ Web search confirms a Zhang 2025 Nat Commun paper on Pou4f3WT/Q113* base editing with Anc80L65 SchABE8e, near-complete hearing recovery, but PMID 40968144 could not be confirmed via direct PubMed lookup; the actual PMID for this paper may differ slightly | ⚠ PAPER LIKELY REAL but PMID not independently confirmed; value (near-complete recovery) matches search results |
| Kim 2023 ACBE, 35–45% A→C | 35% average, up to 45% cell culture | ”Nature Biotechnology, PMID 37322276” | ✅ PAPER EXISTS — “Adenine transversion editors enable precise, efficient A•T-to-C•G base editing,” Nat Biotechnol 2024 (epub 2023), PMID 37322276 confirmed. The paper reports up to 73% with evolved variants. The 35% “average” is more conservative than the actual published data. | ✅ CONFIRMED (note: value is under-stated vs published 73%; conservative is fine) |
| STRC exon coordinate GRCh38/GRCh37 | chr15:43600551 / chr15:43892749 | Standard coordinate liftover | ✅ Plausible given typical chr15 liftover offset (~292 kb) | ✅ |
| Realistic OHC PE efficiency range | 15–40% | “extrapolating from cortical neurons (42%) and cardiomyocytes (35%)” | ✅ Explicitly labeled as extrapolation with stated basis | ✅ APPROPRIATELY DISCLOSED |
| SEED_MISMATCH_GATE | ≥2 seed mismatches required | Standard Cas9 seed-region biology | ✅ Consistent with published Cas9 specificity literature | ✅ |
h07 summary: 7 out of 8 key literature citations confirmed real. 1 confirmed phantom (Gao 2020 PMID 32493791 = COVID paper, replicated from h06). 1 citation likely real but PMID not independently verified (Zhang 2025). The script itself (pe_phase3_5_strcp1_aware.py) has zero external bio-constants — it reads genomic sequence from FASTA and computes mismatch counts. Model risk is in the atomic note prose only.
h08 — STRC ASO Exon Skipping (aso_phase1_design.py, aso_phase2_strcp1_specificity.py)
| Const | Value | Claimed source | Status | Verdict |
|---|---|---|---|---|
| TRANSCRIPT_ID ENST00000450892 | canonical STRC transcript | UniProt Q7RTU9-1, Ensembl | ✅ Verified Ensembl ID | ✅ |
| PROTEIN_ID ENSP00000401513 | STRC canonical protein | translation of ENST00000450892 | ✅ | ✅ |
| E1659A in exon 27 | missense in exon 27 of 29 | stated in atomic note | ⚠ Not verified independently in this audit but consistent with standard exon mapping | ⚠ PLAUSIBLE, not independently checked here |
| RBM24 motifs TGTGTG/GTGTGT, GCTCTTC | ENCODE RBNS ENCSR742AEU, enrichment ≥4.5 | ENCODE accession | ⚠ ENCODE RBNS accession format is valid; web search confirmed RBM24 RBNS data exists on ENCODE portal; specific accession ENCSR742AEU not directly verified (ENCODE portal not queried) | ⚠ ENCODE FORMAT VALID; SPECIFIC ACCESSION UNVERIFIED |
| Nearest-neighbor ΔH/ΔS tables | SantaLucia 1998 DNA:DNA parameters used as proxy for DNA:RNA | Standard SantaLucia NN tables | ✅ The NN values embedded in the script match published SantaLucia 1998 unified parameters exactly (e.g., AA: dH=-7.9, dS=-22.2; CG: dH=-10.6, dS=-27.2) | ✅ CONFIRMED — standard SantaLucia 1998 values, correctly applied |
| FLANK_NT = 80 nt (exon flank context) | 80 nt intronic flank | Standard ASO design convention | ✅ 50–100 nt is standard for splice-blocking ASO design | ✅ |
| ASO_LENS = (18, 20, 22) | Industry standard PMO lengths | ASO chemistry convention | ✅ 18–22mer is the standard PMO range for splice-switching | ✅ |
| DMD precedent: 4 FDA-approved PMOs (Exondys 51, Vyondys 53, Viltepso, Amondys 45) | All 4 named and correctly described | ”FDA-approved” | ✅ All 4 drugs exist, all use PMO chemistry for exon skipping in DMD. Eteplirsen=exon 51 ✅, Golodirsen=exon 53 ✅, Viltolarsen=exon 53 ✅, Casimersen=exon 45 ✅ | ✅ CONFIRMED |
| STRC has 29 exons | ”STRC 29 exons” | Standard annotation | ✅ Correct per Ensembl ENSG00000242866 | ✅ |
| STRC_START/END, STRCP1_START/END chr15 coordinates | STRC: 43.58–43.64 Mb; STRCP1: 43.68–43.72 Mb | ”Verified via PE Phase 4 on-target locus and STRCP1 paralog cluster” | ✅ Consistent with GRCh38 chr15 annotation for both loci | ✅ |
| E1 event: dPSI −0.542, FDR 2.1e-5 | From Sun 2026 rMATS analysis | ”SD03 (Sun 2026)” | ⚠ Sun 2026 is an internal reference to computed splicing analysis; the rbm24_sd03_splicing_analysis.json artifact is real but “Sun 2026” is not a published paper found in web search | ⚠ INTERNAL DATA — the rMATS analysis was done on the project’s own data; “Sun 2026” is mislabelled as a citation if it refers to an internal analysis |
h08 summary: Zero phantom citations in scripts. The ASO Phase 1 script is the most self-contained calculation of the five audited hypotheses — it fetches sequences live from Ensembl REST, uses standard NN thermodynamics with correctly-sourced SantaLucia parameters, and makes no external biophysical claims. Phase 2 script is a pure Hamming-distance scan over chr15 FASTA. Main flag: ENCODE accession ENCSR742AEU not independently verified, and “Sun 2026” is referenced as if a published paper when it appears to be an internal computation.
h27 — STRC STRCP1 Activation Rescue (ohc_strcp1_expression_check.py)
| Const | Value | Claimed source | Status | Verdict |
|---|---|---|---|---|
| GENCODE_STRC | ENSG00000242866.9 | GENCODE v26 as used by GTEx v8 | ✅ Correct GENCODE ID for STRC | ✅ |
| GENCODE_STRCP1 | ENSG00000166763.7 | GENCODE v26 | ✅ Correct GENCODE ID for STRCP1 pseudogene | ✅ |
| GTEx v8 API endpoint | gtexportal.org/api/v2/expression/medianGeneExpression | GTEx public API | ✅ Live API call; real data pulled at runtime | ✅ |
| TPM threshold for “expressed” | > 0.1 TPM | ”standard field convention” | ✅ 0.1 TPM is the widely-used expression floor in GTEx studies | ✅ |
| STRCP1 transcribed in 54 GTEx tissues, TPM 0.01–2.6 | From script output in atomic note | ohc_strcp1_expression_check.json (live GTEx query) | ✅ These values are GTEx-derived and were queried at runtime; reproducible | ✅ |
| STRCP1 GENCODE biotype “transcribed_unprocessed_pseudogene” | GENCODE biotype label | Standard annotation | ✅ Correct per Ensembl/GENCODE annotation | ✅ |
| CRISPRa dCas9-VP64/VPR 10–20× upregulation achievable | Assumed in hypothesis framing | General CRISPRa literature | ⚠ Correct order of magnitude (CRISPRa achieves 2–100× depending on locus/construct) but no specific citation for STRCP1 promoter context | ⚠ PLAUSIBLE — no false number |
| STRCP1 promoter at chr15:~43,700,xxx | Inferred from STRCP1 locus position (~100 kb downstream of STRC) | Genomic logic | ⚠ Position is plausible but not directly fetched; promoter boundaries need Ensembl/GENCODE regulatory annotation lookup | ⚠ ESTIMATE — not computed |
h27 summary: Zero phantom citations. Zero wrong numerical constants. The script is a live API query with no embedded parameters that could be fabricated. This is the cleanest possible audit target: a hypothesis-framing note paired with a live-data-pull script. The only outstanding item is the absence of primary-source backing for the CRISPRa upregulation claim, which is a reasonable background claim rather than a model constant.
Aggregated Root-Cause Analysis
Constants that appear across multiple hypotheses
| Constant | Appears in | Status | Root cause |
|---|---|---|---|
| STRC protein t½ = 14 d | h04, h06 (both scripts and pkpd note) | ⚠ Not primary-lit sourced | Originated in rbm24_mrna_dose_response_results.json as an assumed “dominant eigenvalue”; no stereocilin-specific t½ measurement published |
| RBM24 protein t½ = 2 d | h06 (all three scripts) | ⚠ Not primary-lit sourced | Same JSON source; RBM24 t½ data in OHCs does not exist in literature; 2 d is a plausible heuristic |
| Hill K_M = 200, n = 2, max_boost = 3× | h06 (all scripts) | ⚠ Circular fit to own ODE | Never measured experimentally; values are set so that the model achieves “therapeutic” at a reasonable dose — i.e., fit to a judgment call, not data |
| LNP endosomal escape = 2% | h04, h06 | ⚠ Unsourced generic “literature” | Correct range (published literature consistently <10%) but no specific paper cited; Patel et al. 2019 PNAS (endosomal bottleneck) would anchor this |
| ENDO_MRNA_SS = 50 copies/OHC | h04 only | ❌ Acknowledged biological estimate | No OHC single-cell transcript counting study reports absolute STRC mRNA copies per cell |
| Gao 2020 PMID 32493791 | h06 (atomic note), h07 (atomic note) | 🚨 PHANTOM (= COVID paper) | Both h06 and h07 cite the same wrong PMID for what appears to be an attempt to reference a cochlear LNP-CBE efficiency paper (likely the Gao 2018 Tmc1 CRISPR/lipid paper, PMID 30181527, or a different 2020 study entirely) |
The phantom pattern in batch 2 vs batch 1
Batch 1 (h01, h02, h05, h09, h26): 4/5 hypotheses had phantoms, mostly in external binding affinity constants (Kd, d31, CREB rates).
Batch 2 (h04, h06, h07, h08, h27): 2/5 hypotheses have the Gao 2020 phantom — but notably it is the same phantom in two places, not two independent inventions. This is citation laundering: the wrong PMID (32493791) was generated once and copy-pasted. The underlying claim (LNP cochlear delivery at very low OHC efficiency) is biologically real — the Gao 2018 Beethoven cochlear lipid-delivery paper (PMID 30181527) or the Leclere 2024 review would be appropriate sources — but the PMID is wrong.
The h08 and h27 scripts are structurally different: they make no external bio-constant claims and instead derive values from public APIs or standard tables. These represent the healthiest pattern in the STRC computational portfolio.
Ranking Deltas
| # | Axis change | Verdict |
|---|---|---|
| h04 (B-tier) | No change to tier. All numbers internally consistent; primary-lit gaps are inherited from h06. | B held |
| h06 (A-tier) | 1 confirmed phantom (Gao 2020), 1 wrong-paper-match (Sahin 2014 for ER ceiling), Hill constants circular. Core pharmacology structure sound. | A held but requires: (1) replace Gao 2020 PMID 32493791 with correct cochlear delivery reference; (2) source or acknowledge STRC 14d t½ and RBM24 2d t½ as estimates |
| h07 (C-tier) | Same Gao 2020 phantom in note prose; all other 7 citations confirmed real; scripts have zero constants. | C held (tier is not about citation quality; tier is about delivery feasibility and competition with h03/h04/h08) |
| h08 (C-tier) | No phantoms, no wrong values, cleanest scripts in entire portfolio. | C held (tier reflects scientific risk profile of exon skipping for missense, not citation quality) |
| h27 (C-tier) | No phantoms, no wrong values, hypothesis explicitly pre-Gate-1. | C held pending Gate 1 wet-lab (Holt lab ribosome profiling) |
Recommended Next Actions Per Hypothesis
h04 — Strategy B Full-Length
- Measure or source STRC t½. CHX-pulse-chase in any OHC-like cell line (HEI-OC1 or neonatal mouse organ of Corti) would anchor the 14 d assumption. Alternatively, search for structural extracellular matrix proteins with known t½ in published proteomics turnover datasets (SILAC-based proteome turnover studies). Flag as
ESTIMATED_14Duntil sourced. - Source or replace ENDO_MRNA_SS = 50 copies/OHC. Liu 2020 PNAS single-cell atlas or Ellwanger 2018 Sci Data mouse OHC transcriptome should contain relative STRC expression; converting to absolute copies requires UMI calibration, which most scRNA-seq papers don’t provide. Acceptable to leave as estimate with explicit disclosure.
- No script changes needed.
h06 — Strategy A PKPD
- Replace Gao 2020 PMID 32493791. The correct reference for cochlear LNP delivery efficiency is likely: (a) Gao et al. 2018 PMID 30181527 (Tmc1 CRISPR/lipid neonatal cochlear delivery), or (b) Leclere 2024 PMID 39427589 (review), or (c) pending identification of the actual cochlear CBE + LNP paper.
- Correct Sahin 2014 citation for ER stress 5× ceiling. Sahin 2014 is a drug-discovery review and does not provide a 5× secretory protein UPR threshold. Replace with a primary-literature UPR induction threshold paper or explicitly label ER_CEILING_FOLD = 5.0 as engineering judgment with no primary citation.
- Source STRC t½ and RBM24 t½ (same action as h04).
- Verify Lesperance 2026 Ear and Hearing citation or replace with the confirmed DB-OTO NEJM paper.
h07 — Prime Editing
- Replace Gao 2020 PMID 32493791 in atomic note with correct cochlear delivery reference (same fix as h06).
- Verify Zhang 2025 POU4F3 PMID 40968144 directly on PubMed — the paper appears to exist but the PMID couldn’t be confirmed from web search. Correct if wrong.
- Scripts are clean — no action needed.
h08 — ASO Exon Skipping
- Verify ENCODE accession ENCSR742AEU for RBM24 RBNS data. Direct query to
encodeproject.org/experiments/ENCSR742AEU/would confirm existence. Low priority given it’s a secondary tool-use citation, not a model constant. - Clarify “Sun 2026” attribution — if this refers to the project’s own rMATS analysis of an external dataset, rename to something like “internal splicing analysis (rMATS on Ensembl cochlear RNA-seq, 2026-04-20)” rather than an author-year citation format that implies an external publication.
- Scripts are clean — no action needed.
h27 — STRCP1 Activation Rescue
- No immediate citation actions needed.
- Gate 1 is the only blocking action — email Holt lab re: ribosome profiling on STRC OHC lysates.
- If Gate 1 passes, next audit turn should check CRISPRa upregulation literature for STRCP1 locus-specific data.
Summary: What Is Actually Defensible Today
Defensible as-is (requires no citation fix to run computations):
- h08 ASO Phase 1/2 scripts — purest calculation in the portfolio; Ensembl REST + SantaLucia NN; zero external bio-constants requiring citation
- h27 GTEx expression check — live API query, no embedded constants
- h04 Strategy B model structure — arithmetic is correct; it inherits h06 risk but adds no new phantoms
Requires fix before citing in publications:
- h06 Gao 2020 PMID 32493791 — confirmed phantom; easy fix (find the real cochlear LNP paper)
- h07 Gao 2020 same phantom — copy of same error; fix simultaneously
- h06 Hill constants (K_M, n, max_boost) — not phantoms but circular fits; need either experimental calibration or explicit disclosure as “ODE-fitted parameters with no experimental validation”
Connections
- STRC Cross-Hypothesis Parameter Audit 2026-04-23 — batch 1 (h01, h02, h05, h09, h26)
[see-also]STRC Hypothesis Ranking[see-also]STRC Computational Scripts Inventory[about]Misha