TEMI Learns Molecular Subtypes From WSIs

In a February 9, 2026 PLoS Computational Biology article, Wang, Zhang, and Xiong introduce TEMI, a multimodal framework that trains whole-slide histopathology image models guided by transcriptomic data to classify cancer molecular subtypes. TEMI uses a patch fusion network and aligns WSI representations with embeddings from a masked transcriptomic autoencoder, improving subtype classification and enabling gene expression prediction across multiple cancer cohorts.