Work

Classification of brain tumors using pseudobulk spatial transcriptomics

Year: 2026
Type: preprint
Abstract: Transcriptomic classification of central nervous system (CNS) tumors is emerging as a diagnostic adjunct. However, conventional bulk sequencing requires multiple tissue sections, which is challenging when biopsy material is limited or diffuse infiltration dilutes the neoplastic signal. Spatial transcriptomics can interrogate gene expression from a single tissue section and computationally isolate tumor-enriched regions, but its compatibility with existing bulk-trained transcriptomic classifiers has not been systematically evaluated. We developed a standardized pipeline to generate pseudobulk expression profiles from Visium spatial transcriptomics data and applied three classifiers (a correlation-based k-nearest neighbor [kNN], M&M, and OTTER) to 79 publicly available CNS tumor datasets and 5 institutional Visium HD cases multiplexed onto a single slide. The kNN classifier achieved the highest concordance with reference diagnoses (82.3%), followed by M&M (60.3%) and OTTER (46.5%). Medulloblastoma and ependymoma achieved near-perfect concordance, while high-grade gliomas were the most challenging. Formalin fixation reduced performance, most markedly for OTTER. In cases with low kNN confidence, tumor-cluster enrichment improved concordance from 60% to 76%. Pseudobulk aggregation of spatial transcriptomics data from a single tissue section can therefore yield diagnostically informative classifications using existing classifiers, with potential for tissue-limited specimens and infiltrative tumors.
Language: English
Open Access status: green
OpenAlex ID: W7225918203
Cites: 10
Cited by:
Related to: 10