Integrative Analysis of Spatial Transcriptome and Connectome by SpaCon

Abstract

The brains structure and function arise from its complex molecular composition and neural connectivity, yet the relationship between cell-type-specific gene expression and brain-wide connectivity is not well understood. By integrating single-cell resolution spatial transcriptomics and connectomics, we reveal tight gene-connectivity coupling in the cortico-thalamic circuit. To uncover the latent factors associating connectivity with gene expression, we focused on the often-overlooked extrasomatic mRNAs in spatial transcriptomics and identified specific gene expression in axons of the corpus callosum that reflect their cortical origins. Building on these findings, we developed SpaCon, a deep-learning method that flexibly integrates global connectivity with local gene expression. SpaCon employs efficient neighbor-sampling to enable whole-brain analysis while preserving performance. This architecture allows the model to identify functionally relevant three-dimensional domains defined by transcriptome-connectome patterns, even when spatially distant. Validated across diverse datasets and species, SpaCon significantly enhances the prediction of connectivity from gene expression and improves the spatial classification of neuronal subtypes. SpaCon provides a powerful, scalable, and versatile framework for understanding transcriptome-connectome relationships.