Identification of clinically relevant spatial phenotypes in large-scale multiplex immunofluorescence data via unsupervised graph learning in non-small cell lung cancer

Multiplex immunofluorescence (mIF1) allows simultaneous spatial interrogation of multiple cell- and tissue-based biomarkers from patient cohorts at scale using whole-slide images (WSI).
The study of the spatial relationships between cells is of increasing importance in immuno-oncology. For instance, spatial analyses could inform the effects of a cancer treatment on the tumor immune microenvironment. However, the identification of spatially-derived insights is limited by conventional approaches that reduce spatial data into human-derived feature sets (e.g., nearest neighbor), necessitating new methods for surveying spatial patterns in full.
We hypothesize that an unsupervised approach to mIF analysis using graph neural networks (GNN) will allow identification of ‘spatial phenotypes’ defined by their cellular composition and spatial arrangement in clinical datasets. Here, we identify clinically-relevant, interpretable spatial phenotypes with distinct immunogenic profiles in non-small cell lung cancer (NSCLC).

Egger et al.

Get In Touch