The abstract is read at the publisher; the summary is JClub's.
Kidney-adapted DeepSpot-M virtual spatial transcriptomics doubled gene prediction accuracy (mean Pearson r 0.43 vs 0.22) and preserved disease-marker and eGFR-linked molecular programs in kidney biopsies.
- Why it matters: Spatial transcriptomics is powerful but costly, limiting molecular profiling in clinical diagnostics and large archived biopsies; histology-based prediction could extend compartment-specific programs to existing samples.
- What they did: The model was adapted using paired H&E and Visium data from 209 KPMP participants, then evaluated in held-out human biopsies, in 567 rats with nephrotoxin exposures, and used to build a virtual atlas from 1,534 KPMP slides.
- The result: Virtual profiles preserved directionality of 13 disease-marker differences and all 11 molecular-program–eGFR associations; rat virtual injury scoring distinguished moderate/severe lesions (AUC 0.85) and ≥2-fold blood urea nitrogen (0.82), and atlas program scores linked to eGFR (partial r 0.32) and interstitial fibrosis (-0.57).
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.