GenVinci Rebuilds 3D Tissue From 2D Histology Slides
A preprint describes a tissue foundation model trained on 300 million single-cell and spatial transcriptomic profiles.
Updated on Nov. 18, 2026 in Artificial Intelligence

Researchers introduced GenVinci, a multimodal tissue foundation model that reconstructs 3D tissue scenes from routine 2D histology. The model is built on ImageOmicsNet-300M, a corpus of 300 million single-cell and spatial transcriptomic profiles, posted as a preprint to bioRxiv on November 17, 2026.
Why it matters
Routine 2D histology slides are widely available, so a method that predicts spatial gene expression from them could extend molecular analysis to existing tissue archives. The approach links tissue images to cell-level biology, which may help researchers study aging and disease without requiring separate spatial sequencing for every sample.
ImageOmicsNet-300M pairs 300 million single-cell and spatial transcriptomic profiles with 24.9 million H&E-predicted virtual spatial transcriptomics profiles across 4,623 H&E sections. GenVinci maps senescence-associated states across 85.8 million cells, but the preprint's performance benchmarks against other models are not reported in the extraction.
The details
GenVinci predicts spatial gene expression from H&E images with zero-shot generalization, and its integration with human genetics implicates suprabasal keratinocytes in atopic dermatitis and keratinocyte carcinomas. Suprabasal keratinocytes showed higher p16 and p21 expression in atopic dermatitis and lower expression in keratinocyte carcinomas, according to the researchers.
Timeline
• November 17, 2026: Preprint posted to bioRxiv under DOI 10.64898/2026.11.17.767412.
The Tech Race
GenVinci extends the spatial transcriptomics approach by inferring gene expression from routine H&E slides rather than requiring new sequencing for each sample. The model's reliance on a 300 million profile corpus places it within the growing field of large-scale foundation models applied to biological imaging.
If validated, GenVinci could let researchers estimate cell-level gene activity from the histology slides already stored in hospital and research archives. The preprint is not yet peer reviewed, so the practical tools it could enable remain a research prospect rather than a current clinical option.
The takeaway
Foundation models trained on large paired datasets may let researchers extract molecular information from images that are already routinely collected. Readers should watch for peer review and independent replication before treating the reported findings as established.
Further reading
Explore more Artificial Intelligence coverage to see how foundation models are being applied across scientific research.
Source note: This article includes information reported by Biorxiv.



