CAREER: Predictive spatial omics by graph and generative learning

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

Tissues are social cellular communities. Decoding how distinct cells coordinate their internal molecules and how their interactions give rise to structural tissue shapes is a vital task in informing our understanding of health and disease. Emerging molecular mapping methods have cataloged the chemical maps of individual cells in tissues. The project will develop computerized models of transcript, protein, and metabolite locations across multiple length scales using geometrical rules and data fusion workflows. The project will provide open-source tools for enriching biological insights in tissue images and will create platforms for integrating these concepts as part of immersive educational outreach activities. Opportunities for middle school and high school students, along with the teachers, will be provided to participate in hands-on and digital research in mathematical tissue biology. Single cell, spatially-resolved 'omics methods have revolutionized our understanding of how tissue composition is altered in the progression between states, such as progressing from health to disease. Machine learning methods have been critical to the rapid advances researchers have made in the spatial bioinformatics field. The need for the predictive use of biological maps using graph-based and latent space representation is increasingly recognized as key to our ability to decode tissue structure and function relationships. The goal of the project is to map and interpret hidden tissue features using open-source learning algorithms in image-based spatial-omics data. The specific aims are to 1) Design a cross-scale graph-learning model from subcellular protein interactomics and microstructural tissue topographies for predicting signaling organization and cell communication; 2) Design a cross-modality variational autoencoder model of joint proteo-metabolomics in single cells for dissecting tissue chemical variances in spatial metabolomics and proteomics data. This research will characterize the molecular and structural differences across subcellular and tissue architectures in normal and aberrant phenotypes identified from spatial multi-omics data. The results of the project can be found at the lab website: https://www.coskunlab.org/. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.