This paper introduces MintFlow, a novel generative AI algorithm designed for mapping and reprogramming human tissue microenvironments using spatial transcriptomics data. MintFlow's core innovation is its ability to disentangle intrinsic cellular characteristics from gene expression changes induced by the local microenvironment, moving beyond purely descriptive spatial analysis. Applied to human diseases such as atopic dermatitis, cutaneous melanoma, and clear cell renal cell carcinoma (ccRCC), the model successfully identified spatially organized, disease-specific cell states and microenvironment-induced gene programs (MGPs). Crucially, MintFlow allows for in silico perturbations (virtual simulations) of tissue environments, predicting how cell removal or replacement could reprogram local cellular states and guiding the generation of translational therapeutic hypotheses.
References:
