Poster #20: Pranav Arra

Integrating Spatial Transcriptomics and Metabolomics for Inferring Cell-Cell Interactions

Pranav Arra¹ and Jiekun Yang¹* ¹Department of Genetics, Rutgers University - New Brunswick, New Brunswick, NJ, USA *Corresponding author

Cell-cell interactions are key drivers of tissue function. While ligands and receptors, which can be detected through their respective mRNA expression, play a central role in cell communication, the local metabolic environment may also influence these interactions. Integrating spatial metabolomics and transcriptomics therefore provides an opportunity to investigate how metabolites may modulate or contextualize ligand-receptor interactions within their spatial environment. Here, we propose a metabolite-aware graph learning framework to investigate how local metabolic environments relate to cell-cell interactions. Metabolite m/z ratios and mRNA expression were extracted from a mouse brain sample and aligned into pseudocells, with each pseudocell containing both metabolomic and transcriptomic information. Pseudocells were then grouped into spatially local niches to represent the surrounding cellular environment. Using ligand-receptor interactions inferred from mRNA expression, we use these spatial niches to model the relationship between local metabolite expression and cell-cell interactions with respect to the spatial context. As a proof of concept, we developed a graph convolutional network (GCN) that aggregates metabolite expression across neighboring pseudocells within spatial niches to predict cell-cell interaction scores. The current model infers the relationship between 3,658 metabolites and 50 cell-cell interactions while incorporating the spatial distances between pseudocells. The model currently achieves an average loss of 0.02, corresponding to approximately 5% error, demonstrating its ability to infer relationships between spatially variable metabolites and cell-cell interactions within local environments. Ongoing work will extend this framework to a graph attention network (GAT), where attention scores will provide an interpretable measure of the relationship between metabolites and specific cell-cell interactions. This will allow us to identify which metabolites have the strongest influence on the cell interactions and how the local metabolic environment may modulate cell-cell communication. These relationships can further be used to hypothesize the mechanisms through which metabolites influence cell-cell interactions and other biological processes.