The abstract is read at the publisher; the summary is JClub's.
Loki2, trained on UniSeg’s 15 million segmented cells, recovers single-cell identity and continuous morphology from H&E, with latent dimensions linked to DAPI-validated nuclear phenotypes.
- Why it matters: It addresses whether routine histology holds structured molecular and cellular signals that can be recovered at single-cell resolution, and what biological factors determine that recoverability.
- What they did: Loki2 is a single-cell pathology foundation model trained on UniSeg, then evaluated for cellular representation, DAPI-validated nuclear phenotype links, and transcriptomic/protein-retrieval recoverability using gene-level analyses.
- The result: Molecular recoverability is strongest for genes tied to morphologically distinct cellular identities and functions, and morphology representations are linked to patient staging and biomarker discovery.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.