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

Current single-cell foundation models predominantly borrow architectures from natural language processing, treating genes as unordered tokens and often discretizing expression values. This study introduces scVision, a vision foundation model that converts single-cell transcriptomes into continuous images where genes are placed on a 2D lattice using Gromov-Wasserstein optimal transport based on co-expression patterns. By structuring transcriptomes as images, scVision preserves both expression magnitude and local gene-program co-regulation. Pretrained via masked image modeling on 72 million human cells from the CZ CELLxGENE Census, scVision demonstrates exceptional zero-shot performance across six independent, held-out biological atlases spanning diverse tissues. Across these benchmarks, scVision consistently outperforms existing single-cell foundation models and classical baselines in cell-type annotation, cross-study transfer, and label efficiency—frequently achieving high accuracy with only a single labeled cell per type. Furthermore, scVision's spatial layout provides enhanced interpretability. Analysis of self-attention maps reveals biologically meaningful gene programs across cell types without explicit pathway supervision, uncovering conserved programs such as a p53-linked vascular-immune module across multiple organs. Additionally, scVision matches top integration models in preserving biological structure across studies and exhibits high robustness to gene dropout, establishing vision-based formulations as a powerful paradigm for single-cell biology.

Single-cell transcriptomics has made it possible to measure gene expression in tens of millions of individual cells, revealing cellular diversity that bulk profiling cannot resolve. Foundation models aim to learn general representations from these large datasets that can be reused across many biological tasks. However, most current single-cell foundation models are adapted from language models and represent each cell as a set or sequence of gene tokens. This design has two limitations. It treats genes as largely unordered inputs, even though genes act together in coordinated programs, and it often requires expression values to be discretized or ranked, losing quantitative information about expression magnitude. Here we present scVision, a vision foundation model for single-cell biology. Instead of converting genes into tokens, scVision represents each cell as a continuous gene-expression image. It assigns genes to fixed spatial positions using optimal transport, so that genes with related expression patterns are placed near one another and coordinated gene programs form local image regions. The resulting image preserves both the quantitative expression level of each gene and the biological relationships among genes. We pretrain a vision transformer with masked image modelling on 72 million human cells, creating one of the largest pretrained models for single-cell analysis. In zero-shot evaluations across six independent, held-out studies, frozen scVision representations outperform existing foundation models and classical baselines in cell-type annotation and gene-program discovery, without task-specific retraining. On multi-study integration, scVision matches the strongest token-based foundation model on the combined benchmark score and conserves more biological structure than any method tested. The spatial organization of scVision also improves interpretability: image regions correspond to groups of co-expressed genes, and attention maps can be read as gene-program activity. This structure also enables spatial masking experiments, in which a neighborhood of related genes is perturbed as a single unit, an operation with no direct counterpart in token-based foundation models. By preserving continuous gene-expression values and giving genes biologically meaningful positions, scVision reframes single-cell representation learning as a vision problem. This approach retains more of the original transcriptomic signal while opening a direct path for applying modern computer vision methods to single-cell biology.

The discussion highlights how scVision reframes single-cell representation learning into a vision problem by rendering cell transcriptomes into continuous images using optimal transport. This spatial representation preserves expression magnitude and co-expression relationships, outperforming traditional token-based models in zero-shot cross-study transfer, annotation accuracy, and label efficiency. The authors note limitations such as evaluating only frozen zero-shot embeddings without fine-tuning, training solely on human data, using a single pan-tissue gene layout, and treating attention-derived pathways as exploratory rather than calibrated regulatory inference. Future directions include fine-tuning the backbone, exploring spatial transcriptomics, and multi-modal integration.