SPATIA: Multimodal Generation and Prediction of Spatial Cell Phenotypes

SPATIA is a new multimodal generative and predictive model designed to integrate cellular morphology, gene expression, and spatial context at multiple levels, addressing limitations in existing methods that analyze these modalities in isolation. It employs a confidence-aware flow matching objective for optimal transport reweighting and morphology-profile alignment, enhancing image generation and modeling of phenotypic transitions. Benchmarked against 18 models across 12 tasks, SPATIA demonstrates improved generative fidelity by 8% and predictive accuracy by up to 3%, making it a significant advancement for researchers working with spatial transcriptomics and cellular phenotype modeling.