Featured Project

NeuroPenetrance

A dual-engine platform combining iPSC-derived multi-omics profiling with an AI biomarker layer — built to answer a question genetics still can't: why do two people with the exact same risk variant have such different outcomes?

Incomplete penetrance — genetics' most stubborn unsolved question

Two people can carry the exact same genetic risk variant for a neurodevelopmental disorder — and one develops it, while the other shows no symptoms at all. Clinicians and genetic counselors have no reliable way to predict which outcome to expect. This phenomenon, called incomplete penetrance, is a documented, unresolved problem across clinical genetics — not a niche curiosity, but a real gap affecting families navigating a genetic diagnosis and pharmaceutical trials that fail when they can't identify which patients will actually respond.

Two engines, one compounding advantage

Engine 1 — iPSC Penetrance Profiling

Patient-derived stem cells are differentiated into disease-relevant neurons and profiled using multi-omics data (transcriptomics, proteomics, chromatin accessibility) — capturing how a genetic variant actually behaves in a living neuronal system, not just on paper.

Engine 2 — PenetranSci AI Biomarker Layer

Machine learning models trained on the datasets Engine 1 generates rank the strongest biological predictors of risk, turning raw multi-omics data into an interpretable signal — not a black box.

Every service contract generates proprietary training data that improves the AI layer, which in turn increases the value of each new contract — a compounding advantage that's difficult for a generic lab service or a DNA-only test to replicate.

See the analytical engine in action

This is a real, working prototype — not a mockup. It runs a validated 10-gene immune signature model, built and tested on public patient data (GSE98793), and shows exactly how the model reasons about a sample's risk profile.

Open the demo in a new tab →

From a validated prototype to proprietary iPSC data

This demo runs on public data as a proof of concept. The same architecture — feature ranking, co-expression clustering, permutation-tested classification — will be applied to proprietary iPSC-derived multi-omics data, where the signal is expected to be even stronger and more interpretable.

Discuss a collaboration →