Regulatory Strategy & Causal Safety Audit

Intended Use Statement

The Nilus Lab Discovery Platform is intended solely for Preclinical Target Prioritization and Multi-Cohort Bioinformatic Evidence Auditing in human cardiac ageing and heart failure research.

"The system identifies and audits candidate therapeutic hypotheses across 12 public cardiac datasets, 11 human and 1 mouse (6 single-cell RNA-seq atlases, 3 bulk RNA-seq cohorts, and 3 mass-spectrometry proteomes) combined with Open Targets v4 human genetics, HAGR GenAge/CellAge databases, and IMPC mouse knockout phenotypes."

Regulatory Boundary

Permitted Preclinical RUO

• 12-Dataset Multi-Omics Meta-Analysis
• Intrinsic Cell-Lineage Soup Filtering
• Mouse KO Causal Direction Auditing
• Reproducible Script & JSON Provenance

Prohibited Clinical Use

• Direct Patient Diagnosis
• Therapeutic Prescriptions
• Clinical Decision Support
• Unvalidated Clinical Claims

Status: Research Use Only (RUO). Not a medical device. Not for use in human diagnostic or treatment procedures.

Preclinical Safety Gate

Compensatory Brake & Knockout Directionality Gate

Before any target is kept as a candidate, our audit checks its directional concordance against IMPC Mouse Knockout Phenotypes (Open Targets v4 mousePhenotypes) and published loss-of-function studies to prevent inhibiting endogenous cardioprotective stress brakes:

CONCORDANT (PASS)

Audited Primary Leads

PPP1R12C, TMEM65, SYPL2 & benchmark ATP2A2 remain as computational candidates. ZMAT3, LTBP2 were removed as general ageing genes, CBR4 for protein discordance, COL18A1 for no change with age in GTEx left ventricle, and ZSWIM8 because it is not measurable in human heart protein and its knockout mice have heart defects.

HOLD (PLEIOTROPY)

Lineage Divergence

LGR4 ($H_4 = 0.91$) is placed on hold because cardiomyocyte vs. macrophage knockouts exhibit opposing post-MI remodeling effects.

REJECTED (BRAKES)

Compensatory Shields

ARHGAP1, NOL3 (ARC), and CIRBP are not pursued as inhibitor targets because their rise in heart failure looks protective, not harmful.

Institutional Audit & Validation Roadmap

Completed · Multi-Cohort Discovery & Audit

12-Dataset Discovery & 4-Gate Causal Direction Audit

Completed unblinded discovery across 6 single-cell atlases, 3 bulk cohorts, 3 cardiac proteomes, and Open Targets v4 genetics. Recovered ATP2A2 (SERCA2a) blind (rank 76 of 5,677 genes; 3rd of 11 after all filters), verified 0/15 candidates in HAGR GenAge/CellAge, and eliminated 3 compensatory brakes.

Historical Benchmark Disclosure

Single-Cohort Atrial LODO Clock Deprecation

Leave-one-donor-out evaluation of a 20-D scVI / 100-gene Ridge clock across 54 right-atrial surgical samples (PERIHEART) achieved MAE 6.97 yr, $r = 0.4606$ (recorded in the model file; could not be re-derived from raw data). Single-cohort clock claims were deprecated in favor of the 12-dataset 4-angle discovery architecture.

Next Stage · Wet-Lab Translational Validation

Human iPSC-CM, Cardiac Microvessel & Myofibroblast Assays

Planned, not started: siRNA knockdown of PPP1R12C in unedited iPSC-derived cardiomyocytes, AAV restoration of TMEM65, and a targeted protein assay for SYPL2 in human heart tissue.

Institutional Governance

Data Provenance

All 12 discovery datasets are de-identified, publicly deposited human and mouse reference datasets governed by their respective consortium data use terms (CELLxGENE, GEO, GTEx, PRIDE, Open Targets).

Data Privacy

UK GDPR data-protection alignment is maintained for user authentication metadata.

Regulatory Strategy &
Intended Use