GBM cell-state atlases and the virtual-cell foundation

Author: GlioState Editorial Team | Status: draft, pending review | Last expert review: not yet reviewed (lastReviewed = null) | Version 0.1 | KB snapshot: data/items.csv, audit of 2026-10-16 | A knowledge-base-based narrative review, not a systematic review. Every citation is a verified item in this site's library (VC id) linked to its DOI; preprints are marked and are not peer reviewed.

1. Scope and question

• Question: What do the glioblastoma (GBM) single-cell and spatial atlas items in this library show about malignant cell states and plasticity, and what do they imply for building and evaluating a GBM "virtual cell" (an AI model that predicts how cell states change under perturbation) VC0001?
• In scope: GBM cell-state frameworks and atlases, spatial and plasticity studies, and the general virtual-cell and benchmark literature only as methodological context.
• Out of scope: clinical recommendations; tissue- or patient-scale digital twins and imaging foundation models (planned as separate reviews); spinal cord tumors (the library has no items on them yet).

2. Search and selection log

• Type of review: a narrative review built from the site's knowledge base (KB, data/items.csv). It is not a full systematic search. No external database (PubMed, Europe PMC, bioRxiv, Crossref) was searched for this draft beyond the identifier audit of 2026-10-16; "not found in the KB" is not evidence of absence.
• KB snapshot: 141 items on 2026-10-16; 112 have a DOI/PMID and were checked against Crossref/PubMed in the audit; 29 (news, institutions, conferences, interviews, model pages) have no identifier and are not eligible as evidence.
• KB filter used: (a) `brain_tumor_focus = yes` (49 items), narrowed by hand to peer-reviewed papers, datasets and preprints with a DOI whose title or tags concern cell states, atlases, spatial organization, plasticity, perturbation or GBM-specific virtual-cell attempts; (b) `brain_tumor_focus = no` items with tags such as definition, foundation-model, perturbation or benchmark, kept only as method context. The final selection was an editorial judgment on titles, tags and one-line summaries; no abstracts or full texts were re-read for this draft.
• Flow (counts from the KB only): 141 items in KB, 112 with identifiers, 49 marked brain-tumor-focused, selected for this review: see the reference list for the exact number cited. No de-duplication step was needed beyond treating a preprint and its journal version as one study (see section 7).

3. Inclusion and exclusion criteria

• Included: item is in the KB with a DOI; GBM-specific or other-glioma atlas, spatial or plasticity study; or a GBM-specific virtual-cell/perturbation attempt; or a general virtual-cell/benchmark item used only to frame limits of prediction.
• Excluded: items without an identifier (news, interviews, conference and organization pages); retracted items (none found in the audit); genomic-only studies with no cell-state analysis, such as VC0140 (its KB summary states that the paper has no single-cell state analysis); tissue-scale digital twins, organoid platform papers and imaging foundation models (other planned reviews); a preprint whose journal version is also in the KB is represented by the journal version, e.g. VC0020 is represented by VC0019.
• Non-GBM glioma items (IDH-mutant, H3K27M, pediatric high-grade) are included as comparison and are labelled as not GBM.

4. Evidence summary table

Design and size are taken only from the KB record (title, tags, one-line summary); "not stated in KB" means the KB record does not give it. The proxy grade is the editorial scheme of the series feasibility report (E1 peer-reviewed with original data, E2 peer-reviewed computational or benchmark, E4 preprint). Grades are never upgraded; they are qualitative author judgments, not a systematic score.

4a. GBM and glioma atlas, spatial and plasticity items

4b. General virtual-cell and benchmark context (not GBM-specific)

5. Synthesis by theme

5.1 The Neftel cell-state framework

• The KB item VC0059 reports four malignant GBM states (NPC, OPC, AC and MES, as named in the KB summary), together with plasticity and genetics, from single-cell RNA-seq. This is the state space most later items in the library refer to.
• VC0068 places GBM cells in a normal neurodevelopmental hierarchy, a complementary view in which states are read as positions in a developmental program. The KB does not say how the two views are reconciled (see the verification box).
• Comparison tumors: VC0063 (IDH-mutant glioma lineages), VC0064 (OPC-like programs in H3K27M glioma) and VC0065 (cell states and spatial organization in H3-K27M DMG). VC0141 reports, for IDH-mutant glioma, malignant states that resemble normal glial-neuronal lineage stages or a reactive mesenchymal-like state, mirroring states described in IDH-wild-type GBM. These are not GBM and are not used to draw GBM conclusions.

5.2 GBmap and the atlas layer

• VC0060 (GBmap) is a single-cell and spatial reference atlas of IDH-wild-type GBM with more than 1 million cells, per the KB. It is a resource, not a predictive model.
• VC0069 describes GBM evolution under therapy at single-cell resolution; VC0089 (pediatric high-grade glioma) and VC0141 (IDH-mutant glioma) are longitudinal. VC0090 is a state-resolved atlas of IDH-associated differences in diffuse glioma.
• Within this selection only VC0069 is a GBM treatment-evolution atlas; the other longitudinal items concern other tumor types.

5.3 Spatial architecture and plasticity links

• Spatial structure: GBM has a multi-layered spatial organization VC0061; tumor and host are interdependent in space VC0062; hypoxia coordinates the spatial landscape of myeloid cells and is linked to survival VC0094.
• Preprints (not peer reviewed): niche-specific tumor-cell states from core to edge VC0092; spatial epigenomic niches underlying state plasticity VC0093.
• Plasticity drivers: interactions between cancer and immune cells are reported to drive transitions to mesenchymal-like states VC0066; BRD2 regulates state plasticity and therapy response VC0095.
• Reading: the items connect states to microenvironment and treatment, but the KB records do not allow us to say which links are causal, which are observational, or how consistent they are across cohorts.

5.4 Implications for virtual-cell modeling (editorial interpretation)

• What a virtual cell is meant to do: the AI virtual cell is defined as a multiscale, multimodal model that simulates cell states under perturbation VC0001; a persistent "world model" framing has been proposed VC0131. Single-cell foundation models VC0008, VC0009 and perturbation models VC0016, VC0019 supply the general machinery.
• GBM-specific attempts in the KB: scFOCAL/ISOSCELES links GBM states to drug signatures VC0132; DELPHAI (preprint) targets GBM organoid drug response VC0129; in vivo Perturb-seq gives perturbation-to-state data VC0070; Boolean-network attractors treat states as stable regimes VC0072.
• Caution from benchmarks: deep models have not reliably beaten simple linear baselines VC0048, VC0049; zero-shot performance of foundation models is limited VC0050; zero-shot generalization across contexts is the focus of the Virtual Cell Challenge VC0058. These benchmarks are not on GBM.
• Interpretation: (i) the atlases supply state labels and reference data, but the KB contains few paired perturbation-to-state datasets for GBM (VC0070 is the only in vivo Perturb-seq item); (ii) because states are linked to microenvironment and therapy, a model of tumor cells alone may be insufficient; (iii) any GBM claim should be compared with simple baselines; (iv) the KB contains no peer-reviewed study that applies a general foundation model to GBM state transitions (a finding about this KB and the 2026-10-16 audit, not proof of absence).

6. Evidence grading

Grades in section 4 are carried over from the series feasibility report and are never upgraded in this draft. Items with a published erratum keep their grade and are flagged. Preprints stay at E4 even where a journal version exists elsewhere, until a human re-verifies them.

7. Conflicts, corrections and updates noted

• VC0066 (Hara et al.): the library entry previously pointed to a 2025 Cancer Cell erratum notice; on 2026-10-16 it was corrected to the 2021 original (DOI 10.1016/j.ccell.2021.05.002). The erratum (DOI 10.1016/j.ccell.2025.05.007) is recorded in the item's verification notes. Verify the impact of the erratum on the conclusions cited here. Evidence level not raised.
• VC0068: erratum published (Nat Commun, DOI 10.1038/s41467-020-17979-8); verify impact on cited conclusions.
• VC0016: erratum published (Nat Methods, DOI 10.1038/s41592-023-02073-w); verify impact on cited conclusions.
• Preprint now published: VC0020 (State preprint) has a journal version, which is VC0019; only the journal version is used as evidence.
• Author fix: the author field of VC0058 was normalized to Adduri AK et al. on 2026-10-16.
• Retractions: none found in the 2026-10-16 audit of the 112 items with identifiers. Items without identifiers were not checked.
• Competing perspectives: promising foundation-model claims VC0019 versus benchmarks showing limits VC0048, VC0050. VC0019 and VC0058 come from the same group (Arc Institute, per the KB), a possible conflict to keep in mind when reading the benchmark.

8. Limitations

1. Narrative, KB-based selection; not systematic; selection bias from manual curation; no search-date flow diagram beyond the KB audit.
2. GBM-specific virtual-cell evidence is thin; most virtual-cell items are not on GBM.
3. KB records give one-line summaries; designs and cohort sizes are mostly not stated, and full texts were not re-read for this draft.
4. Several items are preprints (not peer reviewed): VC0092, VC0093, VC0129.
5. Three cited items have errata whose impact is unchecked.
6. Evidence grades are qualitative proxies; no independent replication analysis was done.
7. No prospective clinical validation of any cited virtual-cell approach was identified in the KB.
8. The KB has no date-added field; recency cannot be assessed from it.

9. Clinical commentary

[Placeholder] To be written by the author. Not yet written; the review must not be published as reviewed until it is.

10. References (KB items only)

• VC0001 Bunne C et al.. How to build the virtual cell with artificial intelligence: Priorities and opportunities. Cell, 2024. DOI 10.1016/j.cell.2024.11.015
• VC0008 Cui H, Wang C, Maan H, Pang K, Luo F, Duan N, Wang B.. scGPT: toward building a foundation model for single-cell multi-omics using generative AI. Nature methods, 2024. DOI 10.1038/s41592-024-02201-0
• VC0009 Theodoris CV, Xiao L, Chopra A, Chaffin MD, Al Sayed ZR, Hill MC, Mantineo H, Brydon EM, Zeng Z, Liu XS, Ellinor PT.. Transfer learning enables predictions in network biology. Nature, 2023. DOI 10.1038/s41586-023-06139-9
• VC0016 Bunne C, Stark SG, Gut G, Del Castillo JS, Levesque M, Lehmann KV, Pelkmans L, Krause A, Rätsch G.. Learning single-cell perturbation responses using neural optimal transport. Nature methods, 2023. DOI 10.1038/s41592-023-01969-x — ERRATUM
• VC0019 Adduri AK et al.. Predicting cellular responses to perturbation across diverse contexts with State. Cell, 2026. DOI 10.1016/j.cell.2026.07.052
• VC0020 Adduri AK et al.. Predicting cellular responses to perturbation across diverse contexts with State bioRxiv, 2025. DOI 10.1101/2025.06.26.661135 — PREPRINT
• VC0048 Ahlmann-Eltze C, Huber W, Anders S.. Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines. Nature methods, 2025. DOI 10.1038/s41592-025-02772-6
• VC0049 Csendes G, Sanz G, Szalay KZ, Szalai B.. Benchmarking foundation cell models for post-perturbation RNA-seq prediction. BMC genomics, 2025. DOI 10.1186/s12864-025-11600-2
• VC0050 Kedzierska KZ, Crawford L, Amini AP, Lu AX.. Zero-shot evaluation reveals limitations of single-cell foundation models. Genome biology, 2025. DOI 10.1186/s13059-025-03574-x
• VC0058 Adduri AK et al.. Virtual Cell Challenge 2026: Benchmarking zero-shot generalization across cellular contexts. Cell, 2026. DOI 10.1016/j.cell.2026.08.004
• VC0059 Neftel C et al.. An Integrative Model of Cellular States, Plasticity, and Genetics for Glioblastoma. Cell, 2019. DOI 10.1016/j.cell.2019.06.024
• VC0060 Ruiz-Moreno C et al.. Charting the single-cell and spatial landscape of IDH-wild-type glioblastoma with GBmap. Neuro-oncology, 2025. DOI 10.1093/neuonc/noaf113
• VC0061 Greenwald AC et al.. Integrative spatial analysis reveals a multi-layered organization of glioblastoma. Cell, 2024. DOI 10.1016/j.cell.2024.03.029
• VC0062 Ravi VM et al.. Spatially resolved multi-omics deciphers bidirectional tumor-host interdependence in glioblastoma. Cancer cell, 2022. DOI 10.1016/j.ccell.2022.05.009
• VC0063 Venteicher AS et al.. Decoupling genetics, lineages, and microenvironment in IDH-mutant gliomas by single-cell RNA-seq. Science (New York, N.Y.), 2017. DOI 10.1126/science.aai8478
• VC0064 Filbin MG et al.. Developmental and oncogenic programs in H3K27M gliomas dissected by single-cell RNA-seq. Science (New York, N.Y.), 2018. DOI 10.1126/science.aao4750
• VC0065 Liu I et al.. The landscape of tumor cell states and spatial organization in H3-K27M mutant diffuse midline glioma across age and location. Nature genetics, 2022. DOI 10.1038/s41588-022-01236-3
• VC0066 Hara T et al.. Interactions between cancer cells and immune cells drive transitions to mesenchymal-like states in glioblastoma. Cancer cell, 2021. DOI 10.1016/j.ccell.2021.05.002
• VC0068 Couturier CP et al.. Single-cell RNA-seq reveals that glioblastoma recapitulates a normal neurodevelopmental hierarchy. Nature communications, 2020. DOI 10.1038/s41467-020-17186-5 — ERRATUM
• VC0069 Wang L, Jung J, Babikir H, Shamardani K, Jain S, Feng X, Gupta N, Rosi S, Chang S, Raleigh D, Solomon D, Phillips JJ, Diaz AA.. A single-cell atlas of glioblastoma evolution under therapy reveals cell-intrinsic and cell-extrinsic therapeutic targets. Nature cancer, 2022. DOI 10.1038/s43018-022-00475-x
• VC0070 Liu SJ et al.. In vivo perturb-seq of cancer and microenvironment cells dissects oncologic drivers and radiotherapy responses in glioblastoma. Genome biology, 2024. DOI 10.1186/s13059-024-03404-6
• VC0072 Vieira Junior MG, de Almeida Côrtes AM, Gonçalves Carneiro FR, Carels N, Silva FABD.. A method for in silico exploration of potential glioblastoma multiforme attractors using single-cell RNA sequencing. Scientific reports, 2024. DOI 10.1038/s41598-024-74985-2
• VC0089 Sussman JH et al.. A longitudinal single-cell and spatial multiomic atlas of pediatric high-grade glioma. Cell reports. Medicine, 2026. DOI 10.1016/j.xcrm.2026.102766
• VC0090 Zaman A et al.. A State-Resolved Single-Cell and Spatial Atlas Reveals IDH-Associated Differences in Malignant and Immune Architecture in Diffuse Glioma. MedComm, 2026. DOI 10.1002/mco2.71022
• VC0092 Khan SM et al.. Mapping the spatial architecture of glioblastoma from core to edge delineates niche-specific tumor cell states and intercellular interactions bioRxiv, 2025. DOI 10.1101/2025.04.04.647096 — PREPRINT
• VC0093 Kint S et al.. Spatial epigenomic niches underlie glioblastoma cell state plasticity bioRxiv, 2025. DOI 10.1101/2025.05.09.653178 — PREPRINT
• VC0094 Haley MJ et al.. Hypoxia coordinates the spatial landscape of myeloid cells within glioblastoma to affect survival. Science advances, 2024. DOI 10.1126/sciadv.adj3301
• VC0095 Vadla R et al.. BRD2 bromodomain-mediated regulation of cell state plasticity modulates therapy response in glioblastoma. Neuro-oncology, 2025. DOI 10.1093/neuonc/noaf169
• VC0129 Peng et al.. DELPHAI, AI Agent for Predicting Drug Response and Resistance preprint, 2025. DOI 10.1101/2025.09.12.675826 — PREPRINT
• VC0131 Xing et al.. A world model of the virtual cell Cell, 2026. DOI 10.1016/j.cell.2026.08.042
• VC0132 Suter et al.. Drug and single-cell gene expression integration identifies sensitive and resistant glioblastoma cell populations Nature Communications, 2026. DOI 10.1038/s41467-025-67783-5
• VC0140 Chowdhury T, Tesileanu CMS, Kocakavuk E, Johnson KC, Lee JH, Erson-Omay Z, Heo C, Aldape K, ... Varn FS, Verhaak RGW; GLASS Consortium.. Divergent genomic evolution in astrocytomas and oligodendrogliomas Nature, 2026. DOI 10.1038/s41586-026-11150-x
• VC0141 Johnson KC, Spitzer A, Varn FS, Nomura M, Garofano L, Chowdhury T, Lipsa A, Zhang L, ... Suvà ML, Iavarone A, Tirosh I, Verhaak RGW (51 authors; first five contributed equally). Acquired genetic and cell-state changes in IDH-mutant glioma progression Nature, 2026. DOI 10.1038/s41586-026-10612-6

Disclaimer

For research and educational purposes only. This is not medical advice and not a recommendation of any diagnostic or therapeutic approach, product, company or investment. Patients should consult their own physicians.