SpaConTDS Integrates Multimodal Data For Spatial Domains

Xu et al. publish SpaConTDS on January 29, 2026 in PLoS Computational Biology, a multimodal contrastive learning framework that combines self-supervised contrastive learning with reinforcement learning and a tuple perturbation strategy to identify spatial domains in spatial transcriptomics. The method achieves state-of-the-art domain identification across platforms and resolutions, enables alignment-free multi-slice integration and batch correction, and improves downstream tasks like denoising and trajectory inference.