Scientific
Evidence Registry

Complete empirical audit of Nilus Lab's 4-angle cardiac target discovery pipeline across 6 human single-cell atlases (Reichart 2022, PERIHEART, CAREBANK, Kanemaru, Koenig, Litviňuková), 3 bulk RNA-seq cohorts (MAGNet, GTEx LV/AA), 3 mass-spectrometry cardiac proteomes (Malecki 2025 human LV: 12 young vs 13 older donors, 25 hearts; GTEx: 7 LV + 11 AA donors; Takasugi 2024 mouse heart), and Open Targets v4 human genetics. All targets are at the computational identification stage; none has been tested in the lab.

Unblinded Clinical Benchmark

ATP2A2 Control

Unblinded scan across 5,677 genes recovered positive control ATP2A2 (SERCA2a) blind at rank 76 of 5,677 genes (3rd of the 11 genes left after all filters; $Z_{\text{DCM}} = -7.994$, Malecki human LV protein log₂FC = −0.479, $p = 0.00039$, $q = 0.066$) and candidate TMEM65 at rank 74 of 5,677 ($Z_{\text{DCM}} = -8.008$).

Novelty Audit

0 / 15 in GenAge

0 of 15 candidate genes appear in HAGR GenAge Human (307 genes), GenAge Models (2,205 entries), or CellAge (948 genes). Note that absence from curated databases is only weak evidence of novelty.

Causal Direction Gate

Portfolio Filters

Eliminated 3 compensatory brakes (ARHGAP1, NOL3, CIRBP). Removed from candidates: ZMAT3, LTBP2, COL18A1, CBR4, and ZSWIM8 (not measurable in human heart proteomes; knockout mice have heart defects).

Master Evidence Table

Audited Discovery Candidates & Controls

Every number traces to a named public dataset and an analysis script. The full audit trail is available to partners on request.

Gene	Angle & Compartment	Multi-Cohort Z / Effect	Malecki Human LV Proteome (25 hearts)	GenAge / CellAge	EuropePMC (Heart / Age)	Causal Direction & Tier Verdict
PPP1R12C MBS85 · Candidate	Angle 1 · Mural / Pericyte (1.73×)	MAGNet DCM q = 2.4×10⁻⁵ Reichart DCM q = 0.13	log₂FC = +0.534 (p = 1.1×10⁻⁴) FDR q = 0.037 (main test only)	0 / 0 / 0	2 / 0	CANDIDATE · TARGETED INHIBITION PP1 phosphatase holoenzyme scaffold. Contains AAVS1 locus in intron 1 (Bhagwan et al., F1000Research 2020); requires unedited iPSCs. Computational identification only.
ZSWIM8 Removed	Angle 2B · Not Measurable in Heart MS	Z_RNA = +0.58 (flat) Human heart protein: not measurable	Not measurable (quantified in only 7/25 hearts)	0 / 0 / 0	1 / 0	REMOVED · NOT MEASURABLE Not measurable in either human heart proteome (Malecki or GTEx LV). Zswim8 knockout mice have heart defects and die as newborns. Recommendation to inhibit withdrawn.
SYPL2 MG29 · Candidate	Angle 3 · Cardiomyocyte	Reichart SC CM q = 2.0×10⁻⁵ MAGNet DCM q = 0.19	Measurable in only 4/25 hearts (GTEx LV p = 0.25, n = 7)	MG29 (1999)	1 / 0	CANDIDATE · RESTORE In skeletal muscle, loss of MG29 disrupts T-tubule and triad structure (Nishi et al., J Cell Biol 1999); its role in heart is not established. Lower in DCM cardiomyocytes. Note: H₄ = 0.973 is a liver eQTL at a 44-gene locus, not heart evidence. No change with age in GTEx LV (n=431). Computational identification only.
CBR4 mtFAS II	Angle 2 · Cardiomyocyte (1.5×)	Z_DCM = −7.23 Bulk Age p = 0.127 (null)	log₂FC = +0.269 (p = 0.0497 — sign flip)	0 / 0 / 0	1 / 0	DEMOTED · PROTEIN DISCORDANCE Human LV protein increases (+0.269), directly contradicting restore hypothesis; bulk cardiac ageing is null (p=0.127). Disqualified.
ZMAT3 WIG-1	Angle 1 · Fibroblast (0.867×)	FB SC Age Z = +4.81 (q = 0.0056) Bulk Age Meta-Z = +7.94 (inflated)	Not detected in human hearts (0 of 25 donor LV proteomes)	0 / 0 / 0	2 / 1	DEMOTED · SYSTEMIC AGEING Pre-registered SAP v3 GTEx model demonstrates elevated expression in 9 of 10 non-cardiac tissues. Universal p53 ageing marker, not cardiac-specific.
TMEM65 Candidate (Rank 74)	Angle 3 · Inner Mitochondrial Membrane / ICD	Rank 74 of 5,677 genes Z_DCM = −8.008	Malecki LV p = 0.86 (no change) Mouse Prot −0.42 (q = 0.00014)	0 / 0 / 0	6 / 0	CANDIDATE · RESTORE Inner mitochondrial membrane / intercalated disc protein. Cardiac knockdown in mice causes dilated cardiomyopathy (Teng et al., Nat Commun 2022). No change with age in GTEx LV (n=431). Computational identification only.
COL18A1 Removed	Angle 1 · Pericyte Basement Membrane	GTEx LV age: p = 0.991 (slope = −0.000034)	log₂FC = +0.559 (p = 0.048) GTEx LV p = 0.027 (q_BH = 1.0)	0 / 0 / 0	10 / 2	REMOVED · NO LV AGEING Removed from portfolio: does not change with age in GTEx left ventricle (p = 0.991, n = 431).
LTBP2 TGF-β Dock	Angle 1 · Fibroblast (3.1×)	FB SC Age Z = +2.76 Bulk Age Meta-Z = +6.89 (inflated)	log₂FC = +0.585 (p = 0.418, not significant)	0 / 0 / 0	47 / 11	DEMOTED · SYSTEMIC FIBROSIS Pre-registered SAP v3 model demonstrates elevated expression in 6 of 10 non-cardiac tissues. General systemic fibrotic marker rather than heart-specific driver.
ATP2A2 SERCA2a Control	Positive Control · SR Ca²⁺ Pump	Rank 76 of 5,677 genes Z_DCM = −7.994	log₂FC = −0.479 (p = 0.00039, q = 0.066)	0 / 0 / 0	800 / 48	POSITIVE CONTROL · BENCHMARK Unsteered blind recovery of known heart failure benchmark; 3rd of 11 genes left after all filters. No change with age in GTEx LV (n=431).
LGR4 GPR48	Angle 4 · Genetics-First (H₄ = 0.91, HFpEF GWAS)	Z_DCM = −7.57	Low-abundance GPCR	0 / 0 / 0	6 / 6	HOLD · LINEAGE PLEIOTROPY Cardiomyocyte vs. macrophage knockouts show opposing post-MI remodeling effects.
ARHGAP1, NOL3, CIRBP	Angle 1 & 2 · Compensatory Stress Controls	NOL3 Z = +3.94 CIRBP Z = +4.65	NOL3 p = 0.0044 CIRBP p = 0.013	ARHGAP1 in GenAge (#249/#13)	NOL3: 24/0 CIRBP: 14/2	REJECTED · COMPENSATORY BRAKES Nol3 KO causes dilated LV and fibrosis; Cirbp loss worsens ferroptosis; Arhgap1 KO causes premature aging.

12-Dataset Data Provenance

6 Human Single-Cell / Single-Nucleus Atlases (.h5ad): Litviňuková et al. 2020 (Nature; 486,134 cells, 14 healthy donors), Reichart et al. 2022 (Science; 52 DCM vs 18 healthy donors), PERIHEART 2026 (54 RA donors, 392,819 nuclei), CAREBANK 2025 (23 donors), Kanemaru et al. 2023 (Nature; 21 donors), and Koenig et al. 2022 (Nat Cardiovasc Res; 38 LV donors).
3 Bulk RNA-Seq Cohorts: MAGNet (GSE141910; 166 DCM explants and 166 non-failing donor LV), GTEx v8/v10 Left Ventricle (n = 431), and GTEx Atrial Appendage (n = 429).
3 Mass-Spectrometry Cardiac Proteomes: Malecki et al. 2025 (PRIDE PXD050610; 25 human left-ventricle hearts: 12 donors aged 14–25 vs 13 aged 51–65; 3,431 proteins tested), GTEx Human Heart TMT Proteomics (7 left-ventricle + 11 atrial-appendage donors), and Takasugi et al. 2024 (mouse heart aging proteome, 30 vs 6 months).
Human Genetics & Phenotype Databases: Live Open Targets v4 GraphQL API (GWAS colocalisation H₄ ≥ 0.80, tractability, and IMPC mousePhenotypes), HAGR GenAge & CellAge, and EuropePMC REST API.

Historical Clock & Composite Audit Disclosures

Why We Replaced Single-Cohort Atrial Aging Clocks: In leave-one-donor-out cross-validation across the 54-donor PERIHEART right-atrial appendage surgical cohort, a Ridge model achieved MAE 6.97 yr, r = 0.4606 (recorded in the model file; could not be re-derived from raw data).

Multi-Tissue Specificity Audit: Our earlier 16-gene composite (including PTCHD4, EDA2R, and LINC02388) was significant in 9 of 10 non-cardiac GTEx tissues from the same donors and is therefore a systemic rather than cardiac-specific ageing signature. A separate LV-enriched signature led by CSRP2 replicated in MAGNet (n = 161, p = 2.2×10⁻⁴) as a marker of structural stress remodeling.

Corrective Action: All target discovery now requires cell-type-specific single-cell pseudobulk meta-analysis, human LV protein concordance (Malecki 2025), DCM vs. ageing decomposition, and mouse KO causal direction verification.

Governing Discovery & Audit Equations

Audited Discovery Pipeline

Angle 1 / Non-Myocyte Meta-Analysis

Inverse-Variance Fixed-Effects Meta-Z

Combines donor-level sex-adjusted pseudobulk age slopes across Reichart 2022, PERIHEART, CAREBANK, and Kanemaru:

w_k = 1 / SE_k^2, Z_meta = (∑ w_k β_k) / √(∑ w_k)

Ambient Soup Filter / .h5ad Raw Counts

Intrinsic Cell-Lineage Specificity Gate

Eliminates cardiomyocyte ambient RNA contamination in non-myocyte pseudobulk clusters:

PctExpr(g, Lineage) ≥ 10% AND Mean(g, Lineage) / Mean(g, vCM) ≥ 0.50

Angle 3 / Disease vs. Ageing Decomposition

Cross-Cohort DCM Concordance & Synergy

Requires FDR significance and identical directional sign across single-nucleus Reichart 2022 LV and bulk MAGNet LV (451 genes of 5,677):

q_Reichart < 0.05 ∧ q_MAGNet < 0.05 ∧ sign(β_Reichart) == sign(β_MAGNet)

Angle 4 / Causal Genetics Gate

GWAS Colocalisation & Knockout Concordance

Requires Bayesian colocalisation posterior probability H4 ≥ 0.80 and non-compensatory mouse KO direction:

P(H_4 : shared causal variant) ≥ 0.80 ∧ Phenotype(KO) ≠ Pathogenic_Inversion