This study introduces a technique called NePSTA (neuropathology spatial transcriptomic analysis) that uses spatial transcriptomics with graph neural networks (GNN) for automated tumor subtype identification in the central nervous system (CNS). To develop this technique, researchers used small biopsy fragments (5 micrometer paraffin-embedded tissue sections) from 130 participants to train a graph neural network to predict methylation-based subclasses. The tissue samples from the 130 participants underwent spatial transcriptomics profiling. Ritter et al. hypothesized that the composition of the tumor and neural environment form a distinct molecular fingerprint depending on the CNS pathology, and that these fingerprints can be used to predict tumor subclasses. They demonstrated that spatial transcriptomics could be used to compute copy-number variations (CNVs) for all participants which could be used for robust detection of copy-number alterations (CNAs) that are important for diagnosing the specific tumor type. They then compared methylation-based CNV against the inferred CNV detection using spatial transcriptomics and found both methods detected CNV alterations in 81.2% cases, no alterations in 70.3% of cases, gains in 5.5% of cases, and loss in 5.4% of cases. Furthermore, spatial transcriptomics could be used to determine the spatial resolution in the CNV profiles which could help identify the subclonal tissue architecture. Lastly, the authors tested two GNNs—graph isomorphism network (GIN) and generative adversarial network (GAN)—to classify tumor subtype. They found that the GIN model achieved near- perfect metrics with an accuracy of 0.999, precision of 0.999, recall of 0.998 and F1 score of 0.998, while the GAN model achieved an accuracy of 0.995, precision of 0.991, recall of 0.995 and F1 score of 0.993. Overall, the findings from this study reveal the ability of spatial transcriptomics to identify notable chromosomal alterations which are important when distinguishing between tumor types. Taking advantage of spatial transcriptomics and incorporating them into neuropathological workflows can provide a better understanding of tumor microenvironments, subclonal heterogeneity and cellular interactions.