Entities & typed relations

78 entities that the research topics and library items talk about, with 51 typed relations (promotes, suppresses, marks, transitions_to, validated_by, benchmarked_on, uses_dataset, perturbs, associated_with). Every description cites library items; low-confidence relations are flagged for editorial review. All entity pages are drafts pending expert review.

Cell states

Recurrent transcriptional cell states described in glioma single-cell studies.

Astrocyte-like (AC-like) state

One of the four malignant glioblastoma states of Neftel et al. (AC-like, astrocyte-like). In GlioState’s re-analysis of public data, OPC-like and AC-like cells were frequently confused…

1 items · 2 topics · 2 relations

Glial-progenitor-like cells

Couturier et al. compared adult GBM cells with fetal brain cells and found a conserved neural tri-lineage hierarchy centred on glial progenitor-like cells, which contained most cycling…

1 items · 1 topics · 1 relations

Mesenchymal-like (MES-like) state

One of four recurrent malignant cell states of glioblastoma described by Neftel et al. (2019), alongside NPC-like, OPC-like and AC-like. Studies in the library link transitions toward…

2 items · 3 topics · 11 relations

Neural-progenitor-like (NPC-like) state

One of the four malignant glioblastoma states of Neftel et al., resembling neural progenitors. In that study CDK4 amplification favours the NPC-like state.

1 items · 2 topics · 3 relations

Oligodendrocyte-progenitor-like (OPC-like) state

One of the four malignant glioblastoma states of Neftel et al., resembling oligodendrocyte progenitors; PDGFRA amplification favours it in that study. Single-cell studies of H3K27M…

2 items · 3 topics · 4 relations

Genes

Genes and mutations that appear in the library’s cell-state and therapy studies.

BRD2 (bromodomain containing 2)

In patient-derived xenograft models, PTEN loss recruited BRD2 to MES gene promoters; genetic ablation of BRD2 bromodomains reversed the MES transition and increased radiation sensitivity.

1 items · 2 topics · 3 relations

CDK4 (cyclin dependent kinase 4)

In Neftel et al., CDK4 amplification favours the NPC-like glioblastoma state.

0 items · 2 topics · 1 relations

EGFR (epidermal growth factor receptor)

EGFR amplification is among the genetic alterations (with CDK4, PDGFRA and NF1) that shape the frequencies of glioblastoma cell states in Neftel et al.

0 items · 1 topics · 0 relations

H3 K27M mutation

Histone H3 K27M mutation that defines H3K27M gliomas (diffuse midline glioma). Single-cell studies describe developmental and OPC-like programs in these tumors and map their cell states…

2 items · 1 topics · 2 relations

NF1 (neurofibromin 1)

Gene whose alterations favour the MES-like state in the Neftel glioblastoma framework, where state frequencies are shaped by CDK4, EGFR and PDGFRA amplifications and NF1 mutations.

0 items · 1 topics · 1 relations

OLIG2 (oligodendrocyte transcription factor 2)

scFOCAL linked GBM cell states to drug signatures and identified a combination of an OLIG2 inhibitor with Depatux-M as state-specific sensitivity.

1 items · 1 topics · 0 relations

PDGFRA (platelet derived growth factor receptor alpha)

In Neftel et al., PDGFRA amplification favours the OPC-like glioblastoma state.

0 items · 2 topics · 1 relations

PTEN (phosphatase and tensin homolog)

In patient-derived xenograft models, PTEN loss recruited BRD2 to MES gene promoters.

0 items · 1 topics · 1 relations

Pathways & programs

Transcriptional programs, signalling pathways and microenvironment conditions.

AP-1 transcription factor complex

In 86 patient-matched primary–recurrent glioblastoma pairs, the shift to a mesenchymal phenotype at recurrence was reported to be mediated by AP-1.

0 items · 2 topics · 1 relations

Hypoxia

Spatial studies in the library report hypoxia-dependent niches associated with poor prognosis (GBmap), hypoxia coordinating the spatial landscape of myeloid cells in GBM, and cell-state…

1 items · 2 topics · 2 relations

NF-κB signalling

Library topic text associates BRD2/NF-κB signalling with transitions toward MES-like states in glioblastoma; the association is reported from xenograft and cell models.

0 items · 1 topics · 1 relations

Methods & concepts

Assays, computational approaches, evaluation concepts and treatment modalities.

Agent-based modelling

Mechanistic approach in which individual cells are simulated as agents; the library includes a human-interpretable grammar for PhysiCell multicellular models and an agent-based hybrid…

2 items · 0 topics · 0 relations

CRISPR interference (CRISPRi)

Gene-silencing perturbation delivered intracranially in the in vivo Perturb-seq platform in mouse GBM models.

0 items · 2 topics · 0 relations

Digital twin (tumor)

Computational models calibrated to an individual tumor or model system to simulate growth and treatment response; examples in the library are TumorTwin, anatomy-aware simulation of GBM…

4 items · 1 topics · 1 relations

Single-cell foundation model

Large pretrained models on single-cell data such as scGPT, Geneformer and scFoundation. Benchmarks in the library report limited zero-shot performance and no reliable gain over simple…

25 items · 3 topics · 3 relations

In silico perturbation

Computational prediction of the effect of perturbing genes in a trained model, for example Geneformer in silico deletion; a Nature Protocols article describes discovering candidate…

3 items · 0 topics · 1 relations

Simple linear baseline

Simple linear models used as reference points for perturbation prediction. Ahlmann-Eltze et al. report that deep-learning gene perturbation prediction does not yet outperform them, and…

3 items · 3 topics · 2 relations

Mathematical and mechanistic modelling

Classical and multiscale mechanistic models of glioma, including the Swanson proliferation–invasion model, a DTI-based model of spread along white matter tracts, GBM attractor…

5 items · 2 topics · 0 relations

Neftel four-state framework

Framework from Neftel et al. (2019), combining single-cell RNA-seq of 28 tumors with bulk data from 401 TCGA specimens: malignant GBM cells occupy four main states (NPC-like, OPC-like,…

1 items · 4 topics · 1 relations

Neftel meta-modules

Gene-expression signatures that define the four Neftel states; used in GlioState’s re-analysis of public GSE131928 data, where state calls are signature-score calls and not ground truth.

0 items · 1 topics · 4 relations

Organoid drug screening

Library items include an automated, personalised GBM tumor-organoid drug-screening platform, and DELPHAI (preprint) which models temozolomide responses in GBM organoids.

1 items · 1 topics · 1 relations

Perturb-seq

Single-cell perturbation-screening approach. In vivo Perturb-seq with CRISPRi in mouse GBM models silenced drivers of tumor growth and radiotherapy-sensitising genes; genome-scale…

7 items · 3 topics · 1 relations

Perturbation response prediction

Predicting transcriptional responses to genetic or drug perturbations, as done by GEARS, CPA and State. Independent benchmarks report that deep-learning models do not yet reliably…

12 items · 2 topics · 6 relations

Radiotherapy

Treatment modality that appears across the library’s GBM studies: in vivo Perturb-seq in mouse models showed radiotherapy rewires transcriptional responses to genetic perturbations; BRD2…

2 items · 3 topics · 3 relations

Single-cell RNA sequencing (scRNA-seq)

Assay underlying most cell-state studies and atlases in the library, for example the Neftel study of 28 tumors and the GBmap atlas of more than one million cells.

16 items · 2 topics · 0 relations

Spatial transcriptomics

Gene-expression profiling with tissue coordinates. Library studies use it to describe multi-layered organisation, niches and core-to-edge cell states of glioblastoma; GBmap adds…

7 items · 1 topics · 0 relations

Systematic variation (evaluation concept)

Systema shows that systematic variation between perturbed and control cells can lead to overestimated performance of genetic perturbation response predictors.

1 items · 2 topics · 1 relations

Virtual Cell Challenge

Public benchmark competition launched by the Arc Institute as a “Turing test” for virtual cells; the 2026 edition benchmarks zero-shot generalisation across cellular contexts (six unseen…

5 items · 1 topics · 1 relations

Virtual cell world model (VCWM)

A framing that proposes a persistent multimodal “world model” of the virtual cell, in contrast to task-specific predictors, as an operational definition of the virtual cell.

4 items · 1 topics · 2 relations

Zero-shot evaluation

Evaluating pretrained models without task-specific fine-tuning. Kedzierska et al. report limitations of scGPT and Geneformer in this setting, and the Virtual Cell Challenge 2026…

4 items · 2 topics · 2 relations

Drugs

Therapeutic agents that appear in library studies.

Temozolomide (TMZ)

Chemotherapy agent used in glioblastoma. DELPHAI (preprint) models temozolomide responses in GBM organoids; temozolomide-specific single-cell perturbation data remain scarce.

1 items · 3 topics · 2 relations

Models

Computational models and experimental model systems.

AI virtual cell (AIVC)

Model intended to simulate how cell states change under perturbations. Bunne et al. (Cell, 2024) define it as built on multimodal, multiscale, universal cell representations acting as a…

9 items · 1 topics · 2 relations

AIDO.Cell

GenBio AI’s family of dense, transcriptome-scale-context single-cell models, also presented as a general-purpose cell simulator.

2 items · 0 topics · 0 relations

AlphaCell

Genome-wide virtual-cell world model using optimal-transport conditional flow matching for cross-context perturbation dynamics (preprint).

1 items · 0 topics · 1 relations

Cell2Sentence / C2S-Scale

Approach that represents cells as text for large language models; C2S-Scale scales it to 27 billion parameters for single-cell analysis and virtual screening (preprints).

2 items · 0 topics · 0 relations

CellOT

Neural optimal transport approach to learning single-cell perturbation responses.

1 items · 0 topics · 0 relations

CPA (compositional perturbation autoencoder)

Compositional perturbation autoencoder for responses to drugs, doses and combinations.

1 items · 1 topics · 1 relations

DELPHAI

AI agent for predicting drug response and resistance that models temozolomide responses in GBM organoids (preprint).

1 items · 3 topics · 2 relations

Patient-derived GBM organoids

Experimental model system. A patient-derived GBM organoid biobank recapitulates inter- and intra-tumoral heterogeneity; Core2Edge pairs organoids with brain slices to model infiltration;…

5 items · 2 topics · 3 relations

GEARS

Knowledge-graph graph neural network that predicts outcomes of novel multigene perturbations.

1 items · 1 topics · 1 relations

GeneCompass

Knowledge-informed cross-species foundation model for gene regulatory mechanisms (Chinese Academy of Sciences).

1 items · 0 topics · 0 relations

Geneformer

Context-aware transformer for single-cell data enabling in silico deletion and network predictions; a Nature Protocols article describes target discovery with it. Zero-shot evaluation…

5 items · 1 topics · 3 relations

Nicheformer

Foundation model for single-cell and spatial omics (Theis lab, Nature Methods 2025).

1 items · 0 topics · 0 relations

Patient-derived xenograft (PDX) models

Experimental model system used for the BRD2 findings (MES transition reversed by BRD2 bromodomain ablation) and as the basis of anatomy-aware simulation of patient-specific GBM xenografts.

1 items · 1 topics · 1 relations

PerturbLDM

Conditional latent-diffusion model pretrained on Tahoe-100M for single-cell perturbation responses across drugs, doses and cell lines (preprint).

1 items · 0 topics · 1 relations

scBERT

Early large-scale pretrained language model for cell type annotation of single-cell RNA-seq data (preprint).

1 items · 0 topics · 0 relations

scFOCAL / ISOSCELES

Approach that links glioblastoma single-cell states to LINCS L1000 drug signatures to predict state-specific sensitivity and resistance, with in vitro, ex vivo and in vivo validation…

1 items · 4 topics · 2 relations

scFoundation

100-million-parameter foundation model trained on about 50 million human cells (Tsinghua/BioMap).

1 items · 1 topics · 1 relations

scGen

Early variational-autoencoder model for single-cell perturbation response prediction.

1 items · 0 topics · 0 relations

scGPT

Generative pretrained transformer trained on 33 million cells (Bo Wang lab); a spatial continual-pretraining variant (scGPT-spatial) exists as a preprint. Zero-shot evaluation reports…

3 items · 1 topics · 2 relations

State (Arc Institute)

Arc Institute’s virtual cell model: a state-embedding plus transition model for perturbation response across 70 contexts (Cell 2026; earlier bioRxiv preprint).

5 items · 2 topics · 1 relations

TranscriptFormer

Generative cross-species cell atlas model from CZI trained on 112 million cells from 12 species (preprint).

4 items · 0 topics · 0 relations

TumorTwin

Open, modular Python framework for patient-specific tumor digital twins, demonstrated on high-grade glioma growth and radiotherapy response.

1 items · 1 topics · 2 relations

TxPert

Model that uses multiple knowledge graphs to predict transcriptomic perturbation effects.

1 items · 0 topics · 0 relations

Universal Cell Embeddings (UCE)

Cross-species, zero-shot cell-embedding foundation model (preprint; Leskovec/Quake).

1 items · 0 topics · 0 relations

X-Cell (Xaira)

Xaira’s 4.9-billion-parameter diffusion model for perturbation prediction with reported scaling laws (preprint); its claim to surpass linear baselines has not been peer reviewed.

4 items · 1 topics · 2 relations

Datasets

Atlases, perturbation resources and data platforms.

CellFM-datasets

Unified data infrastructure for pretraining single-cell and spatial transcriptomics foundation models.

1 items · 0 topics · 0 relations

CZ CELLxGENE Discover

Single-cell data platform hosting more than 100 million cells, including glioma datasets.

1 items · 1 topics · 0 relations

Patient-derived GBM organoid biobank

Biobank of patient-derived glioblastoma organoids that recapitulates inter- and intra-tumoral heterogeneity (Jacob et al.).

1 items · 1 topics · 1 relations

GBmap

Harmonised single-cell and spatial reference atlas of IDH-wild-type glioblastoma integrating 26 datasets (over 1.1 million cells, 240 patients) with 7 tumor niches; hypoxia-dependent…

1 items · 2 topics · 1 relations

GSE131928 (Neftel et al. GEO series)

GEO series of the Neftel et al. glioblastoma single-cell data: Smart-seq2 (7,930 cells) and 10X (16,201 cells) from 28 tumors; used by GlioState for a cross-platform replication analysis…

0 items · 1 topics · 1 relations

LINCS L1000 drug signatures

Drug-induced expression signature resource that scFOCAL integrates with GBM single-cell states to predict state-specific drug sensitivity.

1 items · 2 topics · 1 relations

Genome-scale Perturb-seq (K562/RPE1)

Genome-scale Perturb-seq data (Replogle et al.) in K562 and RPE1 cells; a main training and benchmark resource for current models, from cell lines rather than brain tumor cells.

1 items · 1 topics · 0 relations

scBaseCount

AI-agent-curated, standardised, auto-updated single-cell data repository from the Arc Institute.

1 items · 0 topics · 0 relations

Tahoe-100M

Single-cell perturbation atlas of 100 million cells covering 1,100 drugs across 50 cancer cell lines (preprint; peer-reviewed in Cell 2026). Not GBM-focused.

3 items · 1 topics · 1 relations

X-Atlas/Orion

Genome-wide Perturb-seq datasets generated by Xaira on a scalable fix-cryopreserve platform for training dose-dependent foundation models.

1 items · 0 topics · 0 relations

Diseases

Tumor types covered by the library.

Machine-readable: /entities.json, /api/entities/<slug>.json, graph.json, graph.jsonld, schema.json.