The article introduces Novae, a graph-based foundation model designed to analyze spatial transcriptomics data across diverse tissues and technologies. Traditional methods often struggle with batch effects and predefined gene panels, but Novae utilizes self-supervised learning and optimal transport to identify consistent spatial domains without these limitations. By employing a graph attention encoder and learnable prototypes, the model can process millions of cells efficiently while maintaining high spatial resolution. Research highlights its ability to integrate with H&E staining and protein assays to reveal complex immune interactions in cancers and architectural shifts in diseased tissues. Ultimately, Novae serves as a scalable, zero-shot tool that enables researchers to discover new biological biomarkers by harmonizing data from various experimental platforms.

References:

• Blampey Q, Benkirane H, Bercovici N, et al. Novae: a graph-based foundation model for spatial transcriptomics data[J]. Nature Methods, 2025, 22(12): 2539-2550.