A Foundational Generative Model for Cross-platform Unified Enhancement of Spatial Transcriptomics

Spatial transcriptomics (ST) platforms are limited by spatial resolution, sensitivity to low expression levels, alignment with tissue structures, and the balance across tissue complexity. Computational enhancement typically targets a single challenge, e.g., super-resolution using hematoxylin and eosin (H&E) images or sensitivity enhancement with single-cell RNA sequencing (scRNA-seq). However, most ignore the interdependence across challenges, yielding biologically inconsistent enhancement. Here we introduce FOCUS, a foundational generative model for unified ST enhancement, conditioned on H&E images, scRNA-seq references, and spatial co-expression priors. With large-scale pretrained encoders, FOCUS uses a modular design for multimodal integration and a cross-challenge coordination strategy to target co-occurring challenges, enabling joint optimization. FOCUS was trained and comprehensively benchmarked on >1.7 million H&E-ST pairs and >5.8 million single-cell profiles, demonstrating state-of-the-art performance across ten ST platforms, on both individual and coupled challenges. The real-world utility and generalizability were validated on a rare suprasellar tumor, papillary craniopharyngioma, and an unseen ST platform (Open-ST) for primary and metastatic head and neck squamous cell carcinoma.