CancerSTFormer enables multi-scale analysis of spot-resolution spatial transcriptomes and dissects gene and immune regulatory responses to targeted therapies

Abstract The growing number of spot-resolution sequencing based spatial transcriptomic (ST) datasets provides an unprecedented opportunity to study multicellular spatial niches driving cancer transitions. However studying niche-level behavior of tumors remains challenging as it requires a multi-scale approach to modeling the spatial niches and the ability to predict possible effects of genetic perturbations on spatial niches. We propose CancerSTFormer, consisting of a pair of spatially aware transcriptomic foundation models to accommodate niche modeling at different length scales. These models, at the 50µm-Local and 250µm-Extended scales, possess unique capabilities to recover ligand-target gene relationships, niche-specific differentially expressed genes, and organ-specific metastasis associated genes in diverse cancer applications. CancerSTFormer can also reveal the regulatory effects of immune-checkpoint blockade therapies, and other targeted therapies, on patients’ tumors through perturbation analysis, and accurately recapitulated perturbation responses from a spatial Perturb-map experiment. By reusing existing spot-resolution ST studies at scale, this tool transforms the vast spot-resolution ST data into a resource for understanding how gene perturbation impact spatial niches in cancer, while also providing ST-driven, gene-based refinement of treatment-resistance and sensitivity signatures derived from existing bulk transcriptomic studies, enhancing signature interpretation. Competing Interest Statement The authors have declared no competing interest. Footnotes Links to package websites appeared wrong. Added a new figure 6 on perturbation experiment prediction.