Mapping phenotypes to spatial transcriptomics reveals disease-associated microenvironments with PhAST
Brown University
Abstract:
Spatially resolved transcriptomics (SRT) enables gene expression profiling across tissue locations, but linking these measurements to clinical phenotypes remains a major challenge. Here, we present PhAST, a scalable semi-supervised framework that integrates bulk RNA-seq, phenotypic data, and SRT to identify phenotype-associated cells and spatial locations. PhAST jointly learns shared gene programs across bulk and spatial data, models their associations with phenotypes, and incorporates spatial regularization to preserve local spatial structure. PhAST supports diverse phenotype types, scales to datasets with millions of spatial locations, and enables phenotype-informed cross-platform gene imputation. Across comprehensive simulations and five disease SRT applications, PhAST outperforms existing methods and reveals disease-associated spatial microenvironments and candidate therapeutic targets. These results demonstrate that PhAST reveals disease-relevant spatial microenvironments, identifies candidate therapeutic targets, and enables cross-platform knowledge transfer. PhAST further supports phenotype-informed gene imputation, providing a framework for translating disease associations into spatially resolved tissue biology.
Biography:
Jingxuan Bao is a Postdoctoral Research Associate at Brown University’s Data Science Institute. His research develops statistical and machine learning methods for integrating multi-modal biomedical data, including genetics, neuroimaging, single-cell, and spatial transcriptomics, to study disease mechanisms and support biomarker discovery.