Genes as a Dish: A Fresh Look at Cells express
Data on gene activity usually looks like a long string of numbers, like a rainbow stripe (spectrum) with no visible connections. In the scVision model, these numbers are turned into a meaningful picture — not a list of ingredients, but a photo of a finished dish. Genes are arranged on a shared 'recipe map': cooperating genes are placed nearby, their activity shown as brightness, like a light measurement (photometry). The result is an image where genetic programs appear as texture, like layers of dough.
A visual AI, trained on 72 million cells, with no prior preparation (only experience with ordinary photos), outperformed its counterparts in cell recognition. In a surprising twist, when genes were randomly shuffled, accuracy plummeted — even more than when the model itself was disabled. The meaningful arrangement, like plating food on a dish, carries more information than even the sharpest algorithm.
Original abstract
as written by the authors arXiv:2607.14163
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.
The genes on the map were placed by an algorithm reminiscent of striving for the perfect table setting: it balances order and chaos, using the concept of entropy.
The tricorder from Star Trek instantly reads cells and makes a diagnosis. scVision takes a step toward that, turning gene analysis into viewing biological 'photographs'.