This article introduces GASTON, an unsupervised deep learning algorithm designed to analyze complex spatial transcriptomics data. By calculating a metric called isodepth, the tool creates a topographic map of tissue slices that functions similarly to elevation on a geographical map. This innovative approach allows researchers to identify spatial domains while simultaneously modeling both smooth gene expression gradients and abrupt changes in cellular composition. The researchers demonstrate that GASTON outperforms existing computational methods in maintaining spatial coherence, particularly in structured tissues like the brain. Furthermore, the algorithm provides a unique coordinate system to study biological shifts in the tumor microenvironment, such as immune activity and metabolic changes. Ultimately, the software offers a more precise way to visualize how cells organize and communicate within their physical landscape.

References:

• Chitra U, Arnold B J, Sarkar H, et al. Mapping the topography of spatial gene expression with interpretable deep learning[J]. Nature Methods, 2025, 22(2): 298-309.