SPATIA: Multimodal Generation and Prediction of Spatial Cell Phenotypes
Researchers have introduced SPATIA, a novel multimodal generative and predictive model designed to unify and analyze cellular morphology, gene expression, and spatial context. This model operates across multiple levels, from individual cells to entire tissues, addressing limitations of existing methods that often analyze these data types in isolation. SPATIA incorporates a spatially conditioned generative framework with confidence-aware optimal transport reweighting and morphology-profile alignment to accurately model target-state morphology distributions and enable biologically meaningful image generation. Tested on a dataset of 25.9 million cell-gene pairs across 17 tissues, SPATIA demonstrated improved performance over 18 benchmark models, enhancing generative fidelity by 8% and predictive accuracy by up to 3%. AI