Explore endothelial cells.
Start with the tissue.

Wälchli glioblastoma: author FACS CD31+/CD45− endothelial compartment, retaining every EC_clusters annotation, including EndoMT, Stem-to-EC, proliferating and mitochondrial states. These states map to Other unless they carry an explicit vascular class. Controls reuse the existing source-defined adult cohort, including age 15. No fetal, AVM or unsorted non-endothelial cells are added from Wälchli.

Bejarano: all 15,246 cells from the final author EC-only object, separated by disease category. Arteries 1/2, Capillaries and Veins map to their corresponding broad classes; Angiogenic, Interferon and Proliferative stay Other. The non-tumor cohort is one epilepsy surgical reference, and the IDH-mutant cohort is one individual. Reported ages are linked for all retained EC participants from Table S1, worksheet (A) Seq. Precise brain regions, treatment status and metastasis primary sites are not annotated here. Non-tumor and metastasis cells were reused by the authors from Bejarano 2024 and are represented only once here. The separate Wälchli/Xie integration used in the paper is not imported.

Xie 2024: all 57,324 author-annotated ECs, exactly matched to the author expression object by individual, condition and barcode. Adjacent non-malignant tissue (29,810 cells; 7 individuals), IDH-mutant lower-grade glioma (21,222 cells; 3 individuals), and GBM (6,292 cells; 3 individuals) retain eight author labels. GBM-EC Type 1/2 and Inflamatory EC map to Other. Expression uses author log1p(CP10K) converted back to CP10K without a second normalization; the released counts slot is not raw counts. The EC-only CSV provides display coordinates; the all-vascular object UMAP is not used. Ages, precise regions and treatment status are unavailable in the released cell metadata. Tumor-adjacent tissue is shown with controls/non-tumor references in the bar selector but retains its distinct label. Some individuals contribute paired peripheral/tumor specimens; dots are anonymous and no paired statistical test is performed.

Nomura 2025 / GBM-CARE: 5,459 author-annotated endothelial nuclei from 109 samples. Primary GBM (2,258 nuclei), first recurrence (2,630), and second recurrence (571) remain separate cohorts. All five author functional states are preserved and map to Other because they do not establish arterial, capillary or venous identity. Coordinates are a recomputed EC-only UMAP, shared across these cohorts, without batch integration. There is no healthy reference cohort. Clinical locations are tumor locations; three nuclei originate from a specimen labeled Spine. Age at initial diagnosis is available privately, but sampling age is not inferred. Nomura and the companion Spitzer paper represent the same cohort and are included once.

Abdelfattah 2022 via GBmap: 1,287 GBmap-annotated endothelial cells, split into 1,031 primary and 256 recurrent GBM cells. The source original-label field is unknown; arterial/capillary/venous identity is unresolved and maps to Other. Primary and recurrent panels share a recomputed EC-only UMAP without batch correction. There is no internal non-tumor cohort: non-malignant cells inside tumor specimens are not healthy controls. Clinical Supplementary Data 1 and GBmap were checked; ages are unavailable. Broad and specific regions describe tumor locations, not verified cortical dissections. CP10K uses all 25,987 features available in this GBmap release. GBmap is the distribution and annotation source; this is one Abdelfattah cohort, not an additional independent GBmap study.

UMAPs retain separate study coordinate frames; Nomura uses a recomputed EC-only embedding. GBM anatomical subregions and treatment status are not supplied by this downloaded metadata, so they are not inferred. Values are CP10K using all available source features in each cohort, with anonymous equal-donor summaries. The feature sets may differ. Donor IDs and original source metadata stay private.