A vision foundation model for single-cell biology via spatial gene cartography

2026-07-14

AI Summary

This work proposes scVision, the first foundation model that reframes single-cell representation learning as a vision task. Departing from prevailing language-model-inspired approaches that treat cells as gene sequences and neglect inter-gene relationships and expression magnitudes, scVision maps genes into a unified, tissue-agnostic spatial layout via optimal transport, positioning co-expressed genes in spatial proximity. This transforms transcriptomic profiles into continuous images preserving both expression levels and biological structure, enabling pretraining of a Vision Transformer through masked image modeling. Remarkably, scVision achieves zero-shot cell-type annotation and gene program discovery without fine-tuning or batch labels. Across six independent studies, it consistently outperforms existing methods in cell annotation, unsupervised gene program identification, and cross-study integration, with spatial permutation experiments confirming that gene positions encode critical biological signals.

Abstract

Most single-cell foundation models are adapted from language models, representing each cell as a sequence of gene tokens. This discards the relationships among genes and often the magnitude of their expression. We present scVision, a vision foundation model that instead renders each cell as a continuous image. Using optimal transport, it places genes at fixed positions on a single shared, pan-tissue layout so that co-expressed genes become spatial neighbours, turning a transcriptome into an image in which gene programs appear as local texture. We pretrain a vision transformer by masked image modelling on 72 million human cells and use the frozen encoder with no fine-tuning. In zero-shot evaluations on six independent, held-out studies, scVision is the most accurate cell-type annotator and recovers gene programs without supervision, ahead of existing foundation models and classical baselines; on multi-study integration it matches the strongest token-based model while conserving the most biological structure, without ever seeing a batch label. Permuting the gene layout with the network fixed sharply lowers accuracy, more than removing the vision transformer itself, showing that biologically meaningful position, not the network, carries the signal. By preserving expression magnitude and gene relationships, scVision reframes single-cell representation learning as a vision problem, connecting it to the mature methods of computer vision.

Problem

• single-cell biology
• foundation model
• gene expression
• spatial gene relationships
• cell representation

Innovation

• vision foundation model
• spatial gene cartography
• optimal transport
• masked image modeling
• zero-shot cell annotation