Abstract The advancement of technologies to measure highly multiplexed spatial data requires the development of scalable methods that can leverage the spatial information. We present MISTy, a flexible, scalable and explainable machine learning framework for extracting interactions from any spatial omics data. MISTy builds multiple views focusing on different spatial or functional contexts to dissect different effects, such as those from direct neighbours versus those from distant cells. MISTy can be applied to different spatially resolved omics data with dozens to thousands of markers, without the need to perform cell-type annotation. We evaluate the performance of MISTy on an in silico dataset and demonstrate its applicability on three breast cancer datasets, two measured by imaging mass cytometry and one by Visium spatial transcriptomics. We show how we can estimate interactions coming from different spatial contexts that we can relate to tumor progression and clinical features. Our analysis also reveals that the estimated interactions in triple negative breast cancer are associated with clinical outcomes which could improve patient stratification. Finally, we demonstrate the flexibility of MISTy to integrate different kinds of views by modeling activities of pathways estimated from gene expression in a spatial context to analyse intercellular signaling.

Explainable multi-view framework for dissecting intercellular signaling from highly multiplexed spatial data