Transplant molecular diagnostics

A second read on every cardiac biopsy.

HeartMend scores acute cellular rejection risk from a locked 5-gene expression panel — CD7, DEF6, UBE2L6, PSMB9, STAT1 — measured directly from the same endomyocardial biopsy tissue your pathology team already collects.

1.00AUC, independent cohort
92%Accuracy, external test
5Genes, locked panel
HeartMend Panel Result
Biopsy GSM48040
Acute Cellular Rejection
Model confidence
0.94
Reference-normalized expression, locked classifier, no re-training on this cohort.
The diagnostic gap

Endomyocardial biopsy grading is the standard — and it has known blind spots.

~30%

Interobserver disagreement

Histologic grading of rejection severity varies meaningfully between experienced cardiac pathologists, particularly at borderline grades.

1–2mm

Sampling error

Rejection can be patchy. A biopsy forceps samples a small fraction of the myocardium, and a normal read doesn't rule out disease elsewhere.

Days

Turnaround pressure

Treatment decisions for suspected rejection often can't wait for a second opinion or repeat biopsy — a molecular read runs alongside routine histology.

How it works

One biopsy core, one locked model, one report.

The gene panel was fixed before this pipeline ever ran — there's no feature selection happening at read time, only measurement and scoring.

01

Biopsy tissue submitted

A portion of the same endomyocardial biopsy core collected for routine histology is submitted for expression profiling — no additional procedure required.

02

Panel expression measured

Expression levels for the five locked genes — CD7, DEF6, UBE2L6, PSMB9, STAT1 — are measured and normalized against reference values.

03

Locked classifier scores the sample

An L2-regularized logistic regression model, trained once and never re-fit to new cohorts, converts the five values into a rejection probability.

No per-patient retraining · same model weights every time
04

Report returned alongside pathology

A molecular rejection score is delivered to the transplant team as a second, independent data point — to be read together with, not in place of, histologic grading.

External validation

Trained on one cohort. Tested, unchanged, on another.

The classifier was fit once on the training cohort, then applied without any re-fitting to a fully independent external cohort — the strictest available test of whether the panel generalizes.

1.00External AUC
92%External accuracy
p=0.0005Permutation test
0 / 13False positives, non-rejection

External test cohort: 25 biopsies (12 rejection / 13 non-rejection), evaluated against a model trained on a separate 43-sample cohort with no shared patients. Significance assessed by permutation testing (n = 2,000 label shuffles) against the fixed model outputs.

The locked panel

Five genes, chosen for what they track in rejecting myocardium.

CD7
T-cell marker

Marks infiltrating T-lymphocytes, the primary effector cells in acute cellular rejection.

DEF6
Immune signaling

Regulates T-cell receptor signaling and activation within infiltrating lymphocyte populations.

UBE2L6
ISG conjugation

Interferon-stimulated gene involved in the ISGylation response seen during active graft inflammation.

PSMB9
Antigen processing

Immunoproteasome subunit upregulated during interferon-driven antigen presentation in rejecting tissue.

STAT1
Interferon response

Core transcription factor downstream of interferon-gamma signaling, elevated across rejection episodes.

For transplant programs

Built to sit alongside your existing biopsy workflow.

HeartMend doesn't replace histologic grading or your pathology team's read. It's designed to run in parallel, using tissue you're already collecting.

  • No extra procedureUses a portion of the same endomyocardial biopsy core already obtained for histology.
  • Locked model, reproducible readsThe same five genes and the same trained classifier are used for every sample — no site-specific retraining.
  • Independently validatedPerformance reported here reflects an external cohort never seen during model training.
  • Read alongside pathology, not instead of itReports are intended as a complementary data point for the transplant team's clinical judgment.

Considering HeartMend for your transplant program?

We're currently working with a small number of transplant centers on pilot evaluations.

Request a pilot