Integrative Machine Learning for Common Fund Spatial Omics

Carnegie-Mellon University, Pittsburgh PA

Investigators

Abstract

PROJECT SUMMARY / ABSTRACT The NIH Common Fund supports transformative programs that generate large-scale datasets crucial for the broad biomedical research community. The emerging spatial transcriptomics in Common Fund programs provide invaluable insights into cellular diversity and tissue organization. However, integrating and analyzing these diverse datasets remains a significant challenge. To address this gap, we propose developing new machine learning methods to analyze spatial transcriptome data from multiple NIH Common Fund programs, in particular, HuBMAP and SenNet. First, we will develop machine learning methods for integrative analysis of multiple spatial transcriptome datasets. This will include creating a scalable, platform-agnostic framework using pretrained single-cell RNA-seq foundation models to integrate diverse spatial transcriptome datasets and a method to analyze spatial gene programs across multiple samples. Second, we will develop machine learning methods to reveal the underlying mechanisms of spatial gene expression. This will involve creating a transformer-inspired model to disentangle intrinsic cellular factors from intercellular interactions and a framework to identify transcription factors modulating spatial gene programs. Collectively, the methods and tools developed in this project will significantly enhance the utility of existing NIH Common Fund spatial transcriptomics datasets and facilitate integration with other related programs. This will drive forward our understanding of spatial gene regulation and its role in health and disease, aligning with the NIH Common Fund’s mission to support transformative research with broad scientific impact.