The field of spatial omics is rapidly advancing, with a focus on understanding the spatial architecture of the tumor microenvironment (TME) and its implications for precision oncology. Recent developments have highlighted the importance of discovering interpretable spatial biomarkers within the TME, which can reveal insights into immune infiltration, tissue modularity, and other key aspects of cancer biology. Machine learning approaches, such as prototypical part networks and graph convolutional networks, are being explored for their potential to identify meaningful associations between cell populations and biomarkers, as well as to classify cancer versus normal samples. These methods have shown promise in capturing complex relationships between genes and identifying reliable biomarkers from high-dimensional RNA-seq data. Noteworthy papers in this area include:

• ProteinPNet, which uses prototypical part networks to discover TME motifs from spatial proteomics data, consistently identifying biologically meaningful prototypes aligned with different tumor subtypes.
• RGE-GCN, which combines feature selection and classification in a single pipeline, achieving higher accuracy and F1-scores than standard tools for RNA-seq based early cancer detection and biomarker discovery.