Spatial Transcriptomics-Guided Alignment Enhances Molecular Profiling in Pathology Foundation Model
Researchers have developed HEIST, a hierarchical graph transformer model designed to analyze spatial transcriptomics and proteomics data. This model represents tissues as hierarchical graphs, capturing both spatial cell relationships and internal gene expression networks. Pretrained on a massive dataset of cells from various organs, HEIST demonstrates generalization to new data types and achieves state-of-the-art performance in tasks like clinical outcome prediction and cell type annotation. Another study introduces STAMP, a framework that uses spatial transcriptomics to guide pathology foundation models, enhancing their ability to infer molecular profiles from histology images by aligning transcriptomic data into functional pathways. AI
IMPACT These models advance AI's capability in biological research, enabling deeper understanding of cellular processes and disease prediction from complex omics data.