INTRODUCTION: The median survival for glioblastoma multiforme (GBM) has only improved by 4.6 months in more than 100 years, underlying the need for innovative methods for understanding this devastating tumor. Spatial transcriptomics is a new method of whole-transcriptome assessment which allows for simultaneous visualization of mRNA expression in situ with histopathology. METHODS: We applied spARC to 25,517 Unique Molecular Identifiers from 10 surgically resected human IDH-mutant astrocytic gliomas (Grade II (n = 3) and Grade III (n = 3)) and GBMs (n = 4). spARC uses manifold learning, graph filters, data diffusion and diffusion filtration to enable discovery of gene-gene relationships, spatial co-localization of genes, and receptor-ligand interactions applicable across a variety of spatial transcriptomic methods. A Neuropathologist annotated whole-slide images blind to our computational analysis and we used in-situ hybridization and immunohistochemistry to demonstrate that spaRC imputation recovery of gene expression is superior to raw expression. RESULTS: We demonstrate that spARC can identify spatially restricted tumor niches of dense tumor, leading edge, infiltration/high tumor burden, hemoglobin, and vasculature with high specificity. These clusters can be identified with genes including CD44, VIM, LDHA (dense tumor); SNAP25, UCHL1, OLMF1 (leading edge); OLIG2, MBP, PDGFRA (Infiltrating/High Tumor Burden). In our assessment of ligand-receptor interactions we found that GBM samples were enriched for SPP1-CD44 signaling. We validated this using GBM TCGA data stratifying patients based on expression of SPP1 and CD44 genes and found that high expression conferred lower survival at 12 months. CONCLUSIONS: Spatial transcriptomics paired with graph-based machine learning methods applied to human GBMs is a powerful technique able to uncover signaling networks relevant to patient survival and may aid in developing new therapeutic strategies.