🔒 ACCESS RESTRICTED

GAPE Analyzer

Interactive A-score analysis with architecture class selection, serial measurement (E3), intervention target modeling (E5), and cohort context. Live demo access restricted pending clinical partnership.

🖼️ DEMO SCREENSHOT AVAILABLE ON REQUEST

The analyzer computes: A-score across 8 architecture classes, detection tier (NORMAL / MARGINAL / DETECTABLE / FLOOR BREACH), architecture commentary, ranked intervention levers with projected impact, serial drift analysis, intervention target solver, and cohort-context percentile placement. All calibration constants and intervention coefficients are held under Patent 64/014,568 and are available only through executed NDA.

Request Access

Full demo access, screenshots, and calibrated results are available to qualified clinicians, researchers, and institutional partners under appropriate NDA.

Contact: hmahaffeyges@gmail.com
GitHub (public): github.com/hmahaffeyges/IAM-Validation
Patents: 64/012,720 & 64/014,568 (provisional, filed March 2026)

Pan-Tissue Screen — Engine 4

What this screen does: Applies a single mean methylation beta to all 8 architecture classes simultaneously, computes the A-score against each class floor, and ranks by clinical priority.

The honest limitation of bulk beta: A mean beta from a standard blood draw (EPIC 850K array or WGBS) is a weighted mixture dominated by immune/hematopoietic cells (~70% of blood cfDNA). The cycling epithelial fraction (colon, lung, cervical — ~12%) and secretory fraction (breast, liver, prostate, pancreas — ~8%) contribute real but diluted signal. Terminal class (neurons, cardiomyocytes — ~0.5%) and stem classes (<1%) are essentially undetectable in bulk blood. Any A-score computed for those classes from bulk beta is a mathematical extrapolation, not a measurement.

This limitation is now empirically confirmed. VAL-038 (April 18, 2026) tested GAPE's tissue-level predicted ΔA against the largest pan-cancer plasma dataset available (Zeng 2026 Nat Cancer, n=1,294, 14 cancer types). Spearman ρ = −0.02 — honest negative confirming the framework's own prediction. The cancers Zeng finds most detectable in plasma (AML 80%, Lung 76%, Prostate 68%) are not the ones with largest architectural ΔA — they are the ones with the highest tumor-fraction shedding into blood. Plasma detection is a shedding-kinetics phenomenon; architecture is a tissue-state phenomenon. They require different analytical treatment.

What bulk beta actually tells you: Immune class A-score is reliable. Cycling and secretory class A-scores are meaningful but carry dilution uncertainty proportional to their cfDNA fraction. Everything below 4% cfDNA should be treated as exploratory only. Tissue-of-origin localization from bulk plasma is the product-defining step — this screen is the before view, not the intended clinical assay.

The validated workflow — plasma → deconvolution → per-tissue A-score:

1. Tissue-of-origin cfDNA deconvolution (VALIDATED). VAL-041 (April 18, 2026) validated the two-step workflow against 10 cancer types using Moss 2018 tissue-of-origin methylation markers: 10/10 correct top-1 localization, mean max ΔA = +0.174. When plasma IS deconvolved, tissue-of-origin is 100% correct and per-class A-scores recover the full architectural signal that bulk beta dilutes.

2. Cell-type-specific methylation panels — targeted bisulfite sequencing enriched for class-specific CpG loci (colon-specific DMRs, breast-specific DMRs) gives a class-specific beta without the full deconvolution pipeline.

3. Size-selected liquid biopsy — cycling and secretory cfDNA fragments have characteristic nucleosomal sizing. Size-selected enrichment can partially enrich for epithelial classes before methylation measurement.

4. Extracellular vesicle methylation — EVs shed from liver, pancreas, and colon carry tissue-specific methylation patterns at higher concentration than free cfDNA for low-shedding tissues.

The multi-class pre-diagnostic signature — VAL-046 capstone result. Even without per-tissue deconvolution, a multi-class signature aggregated across 7 published pre-diagnostic cohorts (Sister Study breast n=2,776; UK Biobank lung n=680; Nurses' Health colorectal n=355; Rotterdam pancreatic n=182; Health ABC any-cancer n=821 and prostate n=240) shows future-cancer participants with mean ΔA = +0.014 above matched cancer-free controls at baseline, detectable 2–5 years before clinical diagnosis across ≥2 architecture classes (immune, secretory, stromal). The per-patient effect is small, but the multi-class vector is the signal that bulk-beta-across-all-8-classes captures correctly for cohort-level risk stratification.

Alzheimer's multi-class drift — VAL-040 generalizes the framework. Nabais 2021 meta-analysis (n=3,424) showed 4 of 8 architecture classes elevated in AD cohorts (terminal, immune, secretory, stromal) with 7/7 tissue-class combinations showing severity gradient (late-stage > early-stage AD). AD is a systemic multi-class phenomenon detectable peripherally, not a localized neurodegenerative event. The pan-tissue screen is the right instrument shape for this class of multi-class signatures.

Bottom line: The pan-tissue screen with bulk beta is a valid exploratory tool for immune, cycling, and secretory classes, and for multi-class signature detection at cohort scale (VAL-046). For the other five classes at individual-patient resolution it is physics-based computation, not clinical measurement. The right clinical experiment is tissue-of-origin deconvolution followed by class-specific A-score computation. That is what VAL-041 validated and what G-2026-P006 requires for the neurodegeneration prediction.

🔒 Pan-Tissue Screen — Access Restricted

Interactive bulk-beta pan-tissue analysis is restricted. The validated workflow (plasma → deconvolution → per-tissue A-score) is demonstrated in the full demo.

Request access: hmahaffeyges@gmail.com

🔒 ACCESS RESTRICTED

Cancer Database

30 TCGA cancer types · matched tumor-normal pairs · per-class A-score progression · 29/30 confirmed at zero free parameters. The database browser is restricted because it exposes the per-class calibrated A-score progression.

🖼️ DEMO SCREENSHOT AVAILABLE ON REQUEST

Covered cancers include: cycling-epithelial (COAD, READ, LUAD, LUSC, BLCA, OV, STAD, CESC, HNSC, KIRC, KIRP, SKCM, THCA, UCEC, ESCA), secretory (BRCA, LIHC, PAAD, PRAD, ACC, PCPG), terminal (LGG, GBM), stromal (MESO, SARC), immune (LAML, DLBCL, THYM), and the stem/pluripotent inversion case (TGCT). Per-class A-scores, ΔA values, and saturation ceilings are proprietary.

Request Access

🔒 ACCESS RESTRICTED

Cell Architecture Database

8 cell architecture classes · class H_min posteriors from G-002 MCMC calibration · maintenance energetics · architectural ceilings. The cell-class database is restricted because it exposes the calibrated H_min posteriors.

🖼️ DEMO SCREENSHOT AVAILABLE ON REQUEST

Covered classes: Cycling Epithelial, Secretory/Glandular, Immune/Hematopoietic, Terminal/Post-Mitotic, Stromal/Connective, Pluripotent Stem, Adult Tissue Stem, Committed Progenitor. Per-class H_min posteriors are derived from 17 MCMC chains (R-hat < 1.001) against Roadmap Epigenomics healthy reference data and are proprietary under Patent 64/014,568.

Request Access

Open Research Problems — GAPE Registry

These are the open physics and methodology problems in the GAPE framework. Resolved problems are included for completeness — they represent completed milestones. Open problems are not gaps or weaknesses — they are the research frontier.

Resolved

Partial / Confirmed

In progress

Open

Priority

Need data now

G-001

Metabolic Sensitivity (n_bio) Absolute Calibration

OPEN

Ordering confirmed (ρ=0.905, p=0.002 vs Seahorse OCR). Absolute values PRELIMINARY pending paired methylation + metabolic perturbation data.

Approach: emcee sampler. G-007 is the formal MCMC run. Requires Seahorse studies with contemporaneous methylation.

G-002

Class Floor MCMC Calibration

✓ RESOLVED

8 class posteriors converged at R-hat < 1.001. Immune class corrected significantly from the Issue 001 calibration value. All class floors are G-002 posteriors; specific values are proprietary.

Approach: emcee, 5 chains, 8×10^5 samples, R-hat < 1.001. 38 reference cell measurements.

G-003

Architecture Floor from DNMT1 Kinetics

OPEN

First-principles check on G-002 / G-003b posteriors by deriving class floors from published DNMT1 kinetics (Km/Vmax). NOTE: G-003b (MCMC calibration of the 4 non-methylation substrate floors) is RESOLVED at R-hat < 1.001, 800,000 samples; specific values are proprietary. The DNMT1 first-principles derivation is the remaining open piece.

Approach: Published DNMT1 Km/Vmax. Compute minimum ATP per division per class.

G-004

Warburg Transition Position Per Class

OPEN

At what A-score does each class-specific metabolic inversion engage? Currently estimated at A≈1.07 universally. Per-class validation needed.

Approach: TCGA metabolomics vs A-score. Lactate/glucose ratio inflection point per class. CPTAC dataset.

G-005

Replication Throughput Ceiling

OPEN

Derive the cycling rate that saturates methylation maintenance from published repair throughput data.

Approach: MMR repair rate vs mismatch frequency. TCGA mutation burden non-linearity by architecture class.

G-006

Cellular Actualization Ceiling Validation

◐ PARTIAL

MCMC posterior converged: t_max = 120.3 ± 7.1 yr from E(a_bio) fit to DunedinPACE (Belsky 2022). Script 25 in repo. Remaining step: validate inflection point against UK Biobank aging cohort longitudinally.

Approach: DunedinPACE (Belsky 2022). UK Biobank aging cohort. Test inflection point vs observed.

G-007

Metabolic Sensitivity MCMC Confirmation

PRIORITY

Float n_bio on [1,100] in MCMC against published methylation data. Posterior should return ~20.9 at baseline.

Approach: emcee sampler. Likelihood: methylation entropy response to ATP perturbation. Seahorse studies with contemporaneous methylation.

G-008

Cancer Floor Breach Prediction

✓ RESOLVED

29/30 confirmed at zero free parameters. 4,304 matched tumor-normal pairs. TGCT inversion predicted and confirmed. Detection threshold A > 1.05 is physics-derived — no cancer data used.

Approach: TCGA 450K methylation arrays. Threshold derived from architecture floor calibration only.

G-009

Single-Cell Validity

OPEN

Framework derived for cell populations. Does A-score have meaning at single-cell resolution?

Approach: sc-WGBS vs bulk ENCODE. Luo 2017, Smallwood 2014.

G-010

Aging Intervention Predictions

OPEN

Predict A-score impact of senolytics, rapamycin, caloric restriction from first principles.

Approach: Published methylation from dasatinib+quercetin datasets. Rapamycin methylation studies. DunedinPACE before/after.

G-011

Cellular Actualization Ceiling from Enzyme Kinetics

PRIORITY

Derive t_max from DNMT1 kinetics rather than empirical fitting. Critical for trajectory quantification.

Approach: DNMT1 repair rates per cell type. N_CpG = 19.6M. Compare to G-006 MCMC posterior.

G-2026-P006

Alzheimer's Pre-Symptomatic Prediction

◐ PARTIAL

Terminal class A-score will show elevation above 1.02 at least 3 years before clinical AD diagnosis in longitudinal cohorts. VAL-040 (April 18, 2026) confirmed multi-class elevation cross-sectionally: Nabais 2021 meta-analysis n=3,424 shows 4/8 architecture classes elevated (terminal, immune, secretory, stromal), 7/7 severity gradient (late > early AD). The specific pre-symptomatic longitudinal claim (3+ yr before diagnosis) awaits ADNI / BDR archived blood methylation. Is AD what premature neuronal aging drift looks like?

Approach: ADNI, BDR cohort (Shireby 2022 n=631), longitudinal blood methylation with AD outcomes.

G-CANINE-001

Canine Architecture Floor Calibration

✓ RESOLVED

VAL-013 confirmed human-derived class floors predict canine cancer signal at near-invariance across 70 My evolutionary divergence. VAL-025 through VAL-028 confirmed same 104 Wang 2020 Labrador retrievers show monotonic aging trajectory across all 5 substrates. VAL-043 extended to 5 canine cancers (n=104): 4/4 predictions confirmed; canine aging r = 0.9995. Framework is species-independent at zero free parameters.

Approach: Wang 2020 Cell Reports (n=104 Labradors). Horvath 2022 mammalian atlas confirmed 40+ species via VAL-034/035/036 vertebrate extension.

G-DECONV-001

Moss 2018 NNLS Tissue-of-Origin Deconvolution Module

OPEN

Production-grade plasma-to-tissue deconvolution engine for the clinical workflow validated in VAL-041 (10/10 top-1 correct localization, mean max ΔA = +0.174). Would enable per-class A-score from bulk plasma without requiring tissue biopsy. VAL-041 proved the workflow using Moss 2018 markers + scipy.optimize.nnls; productionizing it in web.py is deferred per prior agreement until downstream GEO reference matrix (~30 MB) and QC harness (Salas 2018 bounds) are ready.

Approach: ~300-line new module. Requires 30 MB Moss 2018 reference matrix. scipy.optimize.nnls solver. Salas 2018 QC bounds on neutrophil/lymphocyte/monocyte proportions. Raw β-matrix input form.

G-VAL047-REPL

VAL-047 Independent-Cohort Replication on Frozen Methodology

PRIORITY

VAL-047 passed at the individual-patient level with 10-fold CV Cohen's d in the 0.4–0.8 range across three cancer types (GSE51057 breast d=+0.605; GSE51032 breast replication d=+0.379; GSE51032 colorectal d=+0.835). Class-level predictions were pre-specified; CpG subset and CV scheme were developed during analysis. Honest next step: replicate on a truly independent cohort with frozen methodology (no further CpG tuning). Candidate cohorts: Sister Study extension, UK Biobank participants with archived methylation.

Approach: Freeze class-specific CpG panel and directional weighting scheme. Apply unchanged to independent cohort. Report CV d without post-hoc adjustment.

G-CASCADE-38

VAL-038 Plasma cfDNA — Framework Boundary Documented

✓ RESOLVED

VAL-038 tested whether GAPE-predicted tissue-level ΔA rank-correlates with Zeng 2026 (n=1,294, 14 cancer types) observed plasma cfDNA alteration rate. Result: Spearman ρ = -0.02. Honest negative confirming the framework's own prior finding (VAL-002): plasma cfDNA alteration magnitude depends on tumor-specific shedding kinetics, not on tissue-architectural ΔA alone. This is a framework boundary, not a failure — it is structural: GAPE applied to bulk plasma without deconvolution is out-of-scope. VAL-041 validates the correct two-step workflow.

Approach: Clinical claim discipline: never score bulk plasma directly for cancer detection without tissue-of-origin deconvolution as an intermediate step.

G-BASELINE-80

Healthy Baseline 80-Cell Reference Validation

✓ RESOLVED

80-cell healthy-population reference (8 architecture classes × 10 age decades) compiled from Hannum 2013, Horvath 2013, Roadmap 2015, Moss 2018, Lister 2013, Alisch 2012, Adelman 2019, De Jager 2014 / Shireby 2022, Jaiswal 2014. Gives clinicians age-matched A-score percentiles (p10/p25/p50/p75/p90) per class/decade for two-axis readout (age-percentile × tier). CAVEAT: decades 0–9 and 90+ are underrepresented (single-source cohorts — Alisch pediatric, Rotterdam elderly); wider confidence bands in those bins.

Approach: Published HEALTHY_BASELINES.json in Biological_Physics/validation_runs/. Regenerate with larger pediatric and nonagenarian cohorts when available.

G-CASCADE-46

Multi-Class Pre-Diagnostic Signature (VAL-046 Capstone)

● CONFIRMED

VAL-046 confirmed 4/4 predictions: future-cancer participants across 7 cohort-cancer combinations (Sister Study breast n=2,776; UK Biobank lung n=680; Nurses' Health colorectal n=355; Rotterdam pancreatic n=182; Health ABC any-cancer n=821 and prostate n=240) show mean ΔA = +0.014 above matched cancer-free controls at baseline, detectable 2–5 yr pre-diagnosis across ≥2 architecture classes. Cohort-level signal established. Remaining OPEN: individual-patient clinical utility at ΔA=+0.014 effect size requires serial trajectory analysis (drift rate, not baseline position) OR multi-class pre-specified threshold logic OR integration with orthogonal risk factors.

Approach: Serial blood methylation on same patient over time. Pre-specified multi-class threshold logic validated prospectively on a held-out longitudinal cohort.

Human Scenarios — Wellness Screening

🧪

Select a scenario

Choose from 13 wellness, 12 oncology, or 5 canine guided clinical cases.

Test	Cost (US)	What It Measures	Sensitivity for CRC	Sensitivity for Adenoma
Colonoscopy	$1,250–$4,800 avg ~$2,400 uninsured $0 insured (preventive)	Direct visualization of colon mucosa. Gold standard. Allows removal during procedure.	>95%	~75–90% Misses flat adenomas 27%
FIT (fecal immunochemical)	$10–$50 covered by most insurers	Blood in stool. Detects bleeding lesions only.	60–90%	<40% Misses non-bleeding adenomas
Cologuard (stool DNA)	$599 Medicare $502 / 3 years insurer coverage varies	NDRG4 + BMP3 methylation + KRAS mutation + hemoglobin in stool. Statistical model trained on cancer data.	92%	42% Misses most pre-cancerous adenomas
Epi proColon / mSEPT9 (blood)	~$170 FDA approved 2016 Medicare does not cover	SEPT9 gene methylation in plasma cfDNA. Detects cancer cells shedding aberrantly methylated DNA.	48–72%	11% Not designed for adenoma detection
Shield (Guardant, blood)	~$895 FDA approved 2024 Medicare coverage in progress	cfDNA methylation + copy number + fragment features. ML model trained on cancer cases.	83%	13% Pre-cancer detection remains limited
450K / EPIC array (bulk beta)	$200–$400 research / direct-to-consumer not clinically reimbursed	Global mean methylation beta across 450K–850K CpG sites. GAPE current primary input.	Unknown Not validated for CRC screening	Unknown Pre-clinical research only
GAPE-native cycling class panel (does not yet exist)	Est. $150–$250 at scale; one blood draw no bowel prep required	Targeted bisulfite sequencing of 20–50 cycling-class-specific CpG loci from cfDNA, with tissue-of-origin deconvolution. Physics-derived threshold — no cancer training data required.	Unknown Prospective validation needed	Potentially high Measures pre-cancerous entropy not cancer-shed DNA

Tissue / Class	Existing clinical test	Cost	GAPE-native equivalent
Colon (cycling)	Cologuard, Shield, Epi proColon, FIT	$10–$895	Cycling-class cfDNA panel (not yet built)
Cervix (cycling)	FAM19A4/miR124-2 methylation (European guidelines)	~$80–$150	Closest existing analog to GAPE cycling class
Breast (secretory)	BRCA promoter methylation (research), Galleri (multi-cancer)	$200–$950	Secretory-class cfDNA panel (not yet built)
Prostate (secretory)	PSA (protein), GSTP1/RASSF1A methylation (research)	$30–$150 (PSA)	Secretory-class cfDNA panel (not yet built)
Pancreas (secretory)	CA 19-9 (protein, late), EUS (endoscopy)	$50–$2,000	Secretory-class cfDNA panel (not yet built)
Immune (blood)	CBC with differential, flow cytometry	$15–$200	Bulk blood beta gives immune class directly — most reliable class from blood draw
Neuron (terminal)	CSF tau/amyloid, PET imaging	$1,000–$5,000	Neural cfDNA deconvolution (research only — 0.5% of blood cfDNA)

🔒 ACCESS RESTRICTED

Evidence, Citations & Replication Scripts

G-002 methylation posteriors · G-003b non-methyl H_min grid · G-008 29/30 TCGA confirmation · VAL-001 through VAL-047 including the multi-class drift cascade and GSE51057/GSE51032/GSE69914 per-patient validation. Evidence pack and calibrated replication scripts are restricted because they expose the full calibration layer.

🖼️ DEMO SCREENSHOT AVAILABLE ON REQUEST

What the full evidence pack contains: 8 methylation H_min posteriors (G-002, 17 chains), 32-cell non-methyl H_min grid (G-003b, 17 per substrate), DunedinPACE t_max aging ceiling (G-006), 29/30 TCGA forward-prediction confirmation (G-008), 47 validation entries with pre-registered predictions, effect sizes, statistical significance, and honest negatives.

What's public: The framework is available at the repository linked below — GitHub hosts the manuscript PDFs, cascade validation scripts VAL-037 through VAL-047 (with all calibration constants held privately), and the Zenodo DOI archive.

Patents: 64/012,720 (filed March 21, 2026) covers the derivation pathway from Jacobson 1995 entropy functional to the A-score construction. 64/014,568 (filed March 23, 2026) covers 17 domains and 24 architecture classes across cellular, quantum, and semiconductor substrates.

Request Access

GAPE · Before You Begin

This is not a cancer test.

Understanding what GAPE measures — and what it doesn’t — is essential before reading any result.

Existing blood tests for cancer

Tests like Galleri, CA-125, PSA, and ctDNA assays detect fragments of cancer that has already formed — tumor DNA shed into the bloodstream, proteins produced by a tumor, immune responses to established disease. They answer: is there cancer present right now?

These are powerful tools. GAPE is not competing with them. It is measuring something different.

What GAPE measures

Every cell type has a thermodynamic floor — the minimum methylation entropy required to maintain its identity and function. This floor is derived from the physics of healthy cell maintenance, not from cancer data.

GAPE computes an A-score: the ratio of observed methylation entropy to the healthy architecture floor. A score above 1.05 means the cell’s maintenance machinery is departing from what physics says a healthy cell of this type requires.

GAPE is the thermometer of the cell. Not a biopsy. Not a tumor marker. A reading of the cell’s own thermodynamic state against the physics of what healthy looks like.

The key distinction

Existing tests detect what has happened. A tumor has formed and is shedding material.
GAPE measures the thermodynamic state of the cell. It does not detect tumor fragments. It measures whether the cell’s own epigenomic maintenance is operating within the healthy range.

This is why GAPE may signal earlier. And why it cannot confirm cancer: an elevated A-score means a cell type is showing more entropy than healthy physics permits. It does not tell you why.

Pre-clinical research tool only. GAPE results are not a diagnosis, not a screening recommendation, and not a substitute for clinical evaluation. All findings derive from published peer-reviewed sources. Not a substitute for professional medical advice.

Mahaffey (2026) · doi:10.5281/zenodo.19547624 · Patents 64/012,720 & 64/014,568