Concordia: Spatial Domain Detection via Augmented Graphs for Population-Level Spatial Proteomics
Liu, S.; Hsu, L.; Sun, W.
Show abstract
A key step in analyzing population-level spatial proteomic data is to delineate consistently defined spatial domains across samples. Domain detection is particularly challenging for cancer tissues, which have complex spatial domains with elongated or branching geometries. To address these challenges, we present Concordia, a Graph Neural Network (GNN)-based framework that uses augmented graphs to capture complex spatial domains, and it is designed to analyze thousands of tissues simultaneously to obtain consistently defined domains. Applied to a lung cancer dataset, Concordia uncovers a spatially defined cancer associated fibroblast subset linked to clinical outcomes that cannot be identified using protein expression alone.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.