Computational Pathology

Staining,
inferred.

Virtual staining uses machine learning to predict how tissue would look under expensive or unavailable stains, from images that already exist.

01 What it is

One stain, predicted from another.

Pathology runs on stains. Each one takes time, costs money, and consumes tissue. Virtual staining trains a model on pairs of the same tissue stained two ways, so it learns to predict the second stain from the first. Once trained, it produces a predicted stain in seconds, with no reagents and no additional tissue.

Cost
A fraction of the cost
A physical immunostain is expensive in reagent and labor for every slide. A virtual stain costs only the compute at inference, a tiny fraction of that.
Speed
Seconds, not days
A stain that takes a lab hours or days can be predicted in moments, collapsing the iteration cycle for research that depends on many specimens.
Scale
Archives, unlocked
Models can be applied to imagery that already exists, studying biology that physical staining could never reach.

Why it matters

Pathology has always been limited by the stain, not the tissue.

Virtual staining changes what is affordable to see, and with it, what is possible to ask.

02From the journal

Reading on virtual staining.

June 2, 2026
How do you validate a virtual stain?

A predicted stain is only useful if it is trustworthy. How virtual staining models are checked against ground truth, and what validation can and cannot settle.

Read the primer →All writing →
03 Get in touch

Working on this too, or just curious? Say hello.

Email

hello@gradientpath.bio