SpaSEG: unsupervised deep learning for multi-task analysis of spatially resolved transcriptomics

Abstract

Spatially resolved transcriptomics (SRT) for characterizing spatial cellular heterogeneities in tissue environments requires systematic analytical approaches to elucidate gene expression variations within their physiological context. Here, we introduce SpaSEG, an unsupervised deep learning model utilizing convolutional neural networks for multiple SRT analysis tasks. Extensive evaluations across diverse SRT datasets generated by various platforms demonstrate SpaSEG’s superior robustness and efficiency compared to existing methods. In the application analysis of invasive ductal carcinoma, SpaSEG successfully unravels intratumoral heterogeneity and delivers insights into immunoregulatory mechanisms. These results highlight SpaSEG’s substantial potential for exploring tissue architectures and pathological biology.