cerelume

Our science · The problem

A glioma diagnosis is now a molecular diagnosis, and today it takes an operation to get one.

Every treatment guideline keys its recommendations to molecular markers that only tissue can supply, so the molecular answer depends on surgery at every stage of the disease.1,2,3,4 The cost falls on patients: surgery planned without knowing the tumour type, treatment decisions delayed by a month or more,5 repeat operations to find out whether the disease has progressed, and months lost on drugs that have stopped working. For the patient who cannot be operated on there is no specific diagnosis at all, so treatment is chosen without it, trials and the newest drug are closed to them, and median survival is measured in months.6

01

No molecular answer before surgery

A patient with symptoms or a suspicious scan cannot learn what the tumour is without an operation.1,3

The outcome

Surgery is planned without knowing the tumour type, and how aggressive to be is decided blind.

02

The diagnosis arrives a month or more after surgery

In a prospective cohort, a methylation-based diagnosis took a mean of 33 days (range 19 to 81), and 68 days when sequencing was needed.5

The outcome

Treatment decisions wait, in a disease where survival can be measured in months.

03

Some tumours cannot be operated on

Tumours in functional or deep brain regions, and frail or older patients, may have a biopsy only or no surgery at all, so they never get a full molecular diagnosis.2,6

The outcome

Radiotherapy or temozolomide is chosen on age and performance status, not biology. Median survival in biopsy-only elderly glioblastoma is 4.6 months.

04

Locked out of the one disease-modifying drug

The FDA label for vorasidenib in grade 2 glioma requires surgery, including biopsy.7,8

The outcome

A patient who could benefit from an IDH inhibitor is formally ineligible to receive it.

05

No evidence is generated for them

Standard phase III criteria excluded 57% of a real-world glioblastoma population, and having had surgery is itself a criterion.9

The outcome

Their treatment is never tested, so it never improves. The evidence gap sustains itself.

06

Progression and treatment response are judged on MRI alone

Response criteria (RANO 2.0) rely on imaging, and about a third of apparent progressions on MRI turn out to be pseudoprogression.10,11

The outcome

Working treatment is stopped, failing treatment is continued, or the patient faces another operation to find out.

07

Drug resistance is found by failure

Temozolomide and IDH-inhibitor resistance arise through molecular change that cannot be tracked without repeat tissue.12,13

The outcome

Resistance is recognised only when the scan gets worse, after months lost on a drug that had stopped working.

08

In lower-grade glioma, pathologists often disagree

Central review disagreed with the local diagnosis in 43% of 500 routinely reviewed gliomas. Agreement is 96% for glioblastoma but far lower for grade 2 to 3 and mixed histology.14

The outcome

In the grades where the IDH-inhibitor decision sits, the same patient can receive different diagnoses at different centres.

The patient pathway, in numbers

Prevalence

324,095

new brain and CNS cancer cases worldwide in 2022, with 248,964 deaths. It ranks 19th among cancers for new cases but 11th for deaths.15

Diagnosis

33 days

mean wait from surgery to a molecular diagnosis, and 68 days when sequencing is needed. Where molecular testing is unavailable, 24% of gliomas stay unclassified.5,16

Treatment

57%

of a real-world glioblastoma cohort is excluded from phase III trials. Without a molecular result, treatment is chosen on age and performance status.9,2

Prognosis

4.6 months

median survival in biopsy-only elderly glioblastoma, and 1.6 months with poor performance status. One-year survival runs from 98% in grade 2 IDH-mutant astrocytoma to 53.7% in IDH-wildtype glioblastoma: molecular class decides prognosis.6,17

Patients managed on imaging alone never enter surgical series and are not counted in these figures.

Existing liquid biopsies have not solved this.

The blood-brain barrier limits how much tumour DNA reaches the blood. In a study of 640 patients, circulating tumour DNA was detectable in fewer than half of primary brain tumours, against more than three quarters of most advanced cancers,18 so mutation-based tests and multi-cancer early-detection tests perform poorly for glioma. Cerebrospinal fluid tests perform better but need a lumbar puncture and cannot be used for frequent monitoring.19

Our science · The solution

MethyGlio: a liquid biopsy that reports glioma and its molecular class, without surgery.

MethyGlio reads cell-free DNA (cfDNA) methylation with an interpretable deep learning classifier, to diagnose without a surgical procedure and to follow the tumour over time, so that intervention can come earlier and outcomes can improve.20 Methylation is a broad, information-rich signal that reveals tumour identity even when the tumour fraction is faint.21

See the evidence

The largest atlas built for this purpose

MethyGlio is trained on 15,606 harmonised methylation profiles: 6,750 CNS tumours, 7,653 non-CNS tumours and 1,203 normal tissues.20,21

Classifies the way a clinician thinks

A hierarchical Transformer answers six questions in sequence: tumour or not, CNS or not, tumour entity, glioma subtype, grade and molecular class (IDH, 1p/19q, H3 and others).20

Prospectively validated on a different platform

In a prospective cohort of 68 consecutive neurosurgical patients, profiled on newer array and sequencing chemistry, AUC was 0.90 to 0.96 across all classification levels with no retraining (data on file).20

Reaches the patients guidelines cannot

Molecular class from a blood draw at the time of first imaging, CT or MRI: for the patient who cannot be operated on, for grade 2 disease under surveillance, and for centres with imaging but no molecular neuropathology.16

Explainable, and used to design the liquid biopsy

Integrated Gradients shows which CpG sites drive each call. We used those attributions to select loci that survive in blood, then confirmed serum-to-tumour concordance at single-locus resolution in matched patient samples by whole-genome bisulfite sequencing. The panel includes novel proprietary biomarkers.22,20

The guidelines have opened a door.

The 2026 NCCN update is reported to extend cerebrospinal fluid molecular profiling to inoperable high-grade glioma, whatever the reason surgery is not possible.23 That would mean a guideline body accepting the principle of tissue-free molecular diagnosis. What remains uncovered is grade 2 glioma, where the IDH-inhibitor decision sits, patients who cannot have a lumbar puncture, and repeated monitoring. That is where a blood-based liquid biopsy belongs.

What the result changes

Guidelines are decision trees indexed on molecular markers.2,3 The value of a liquid biopsy result is the decision it unlocks, not the label it assigns.

  1. MGMT status

    Temozolomide alone or radiotherapy alone for the older or frail patient, the choice the guidelines make on a marker these patients cannot supply.

  2. IDH and 1p/19q status

    Referral for an IDH inhibitor in grade 2 glioma, and PCV against temozolomide in oligodendroglioma.

  3. Grade and molecular class

    Whether a patient enters a treatment guideline at all, whether a trial is open to them, and whether a biopsy is worth reconsidering.

Evidence, not a black box

Designed for scientific scrutiny.

MethyGlio was evaluated with locked, leakage-controlled testing, then tested on a cohort it had never seen, on a platform it was not trained on. The liquid biopsy assay is at proof-of-concept stage and is not yet validated for clinical use. All MethyGlio figures are data on file while the manuscript is under review.

Atlas scale

15,606 harmonised samples: 6,750 CNS tumours, 7,653 non-CNS tumours and 1,203 normal tissues, drawn from public repositories worldwide.

External validation

68 consecutive neurosurgical patients, enrolled prospectively and profiled on newer array and sequencing chemistry: AUC 0.90 to 0.96 across all classification levels with no retraining (data on file).

Platform independence

Validated across 450K, EPIC and EPIC v2 arrays and whole-genome bisulfite sequencing, so the model can move to a low-cost targeted assay.

Explainability

Every call comes with the CpG sites behind it, which supports clinician trust, regulatory review and rapid assay design.

Collaborate with Cerelume

Help bring a clearer signal to brain tumour care.

Start a conversation

References

  1. 1Louis DN, et al. The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro-Oncology. 2021;23:1231-1251. doi:10.1093/neuonc/noab106
  2. 2Mohile NA, et al. Therapy for Diffuse Astrocytic and Oligodendroglial Tumors in Adults: ASCO-SNO Guideline. J Clin Oncol. 2022;40:403-426. doi:10.1200/JCO.21.02036
  3. 3Weller M, et al. EANO guidelines on the diagnosis and treatment of diffuse gliomas of adulthood. Nat Rev Clin Oncol. 2020;18:170-186. doi:10.1038/s41571-020-00447-z
  4. 4National Institute for Health and Care Excellence. Brain tumours (primary) and brain metastases in over 16s. NICE guideline NG99. 2018, updated 2021. nice.org.uk/guidance/ng99
  5. 5Deacon S, et al. ROBIN: A unified nanopore-based assay integrating intraoperative methylome classification and next-day comprehensive profiling for ultra-rapid tumor diagnosis. Neuro-Oncology. 2025;27:2035-2046. doi:10.1093/neuonc/noaf103
  6. 6Edriss J, et al. Factors associated with poor prognosis in elderly biopsy-only glioblastoma patients. Acta Neurochir. 2025;167:273. doi:10.1007/s00701-025-06681-7
  7. 7US Food and Drug Administration. FDA approves vorasidenib for Grade 2 astrocytoma or oligodendroglioma with a susceptible IDH1 or IDH2 mutation. 6 August 2024. fda.gov
  8. 8Mellinghoff IK, et al. Vorasidenib in IDH1- or IDH2-Mutant Low-Grade Glioma. N Engl J Med. 2023;389:589-601. doi:10.1056/NEJMoa2304194
  9. 9Skaga E, et al. Real-world validity of randomized controlled phase III trials in newly diagnosed glioblastoma: to whom do the results of the trials apply?. Neurooncol Adv. 2021;3:vdab008. doi:10.1093/noajnl/vdab008
  10. 10Wen PY, et al. RANO 2.0: Update to the Response Assessment in Neuro-Oncology Criteria for High- and Low-Grade Gliomas in Adults. J Clin Oncol. 2023;41:5187-5199. doi:10.1200/JCO.23.01059
  11. 11Abbasi AW, et al. Incidence of Tumour Progression and Pseudoprogression in High-Grade Gliomas: a Systematic Review and Meta-Analysis. Clin Neuroradiol. 2017;28:401-411. doi:10.1007/s00062-017-0584-x
  12. 12Touat M, et al. Mechanisms and therapeutic implications of hypermutation in gliomas. Nature. 2020;580:517-523. doi:10.1038/s41586-020-2209-9
  13. 13Taglang C, et al. Understanding and imaging PHGDH-driven intrinsic resistance to mutant IDH inhibition in gliomas. bioRxiv. 2025. Preprint. doi:10.1101/2025.08.30.673205
  14. 14van den Bent MJ. Interobserver variation of the histopathological diagnosis in clinical trials on glioma: a clinician’s perspective. Acta Neuropathol. 2010;120:297-304. doi:10.1007/s00401-010-0725-7
  15. 15Ferlay J, et al. Global Cancer Observatory: Cancer Today. Brain, central nervous system fact sheet (GLOBOCAN 2022). International Agency for Research on Cancer; 2024. gco.iarc.who.int
  16. 16Paolinelli PS, et al. Accessibility to molecular studies in gliomas: a challenge for precision medicine in low-resource countries. Front Oncol. 2026;16:1731276. doi:10.3389/fonc.2026.1731276
  17. 17Ostrom QT, et al. National-level overall survival patterns for molecularly-defined diffuse glioma types in the United States. Neuro-Oncology. 2022;25:799-807. doi:10.1093/neuonc/noac198
  18. 18Bettegowda C, et al. Detection of Circulating Tumor DNA in Early- and Late-Stage Human Malignancies. Sci Transl Med. 2014;6. doi:10.1126/scitranslmed.3007094
  19. 19Otsuji R, et al. Liquid Biopsy for Glioma Using Cell-Free DNA in Cerebrospinal Fluid. Cancers. 2024;16:1009. doi:10.3390/cancers16051009
  20. 20Maimaiti A, et al. MethyGlio: cfDNA methylation and liquid biopsy for glioma classification. Manuscript under journal review. Data on file, Cerelume; 2026.
  21. 21Capper D, et al. DNA methylation-based classification of central nervous system tumours. Nature. 2018;555:469-474. doi:10.1038/nature26000
  22. 22Sundararajan M, Taly A, Yan Q. Axiomatic attribution for deep networks. Proceedings of the 34th International Conference on Machine Learning. PMLR. 2017;70:3319-3328. arXiv:1703.01365
  23. 23National Comprehensive Cancer Network. NCCN Clinical Practice Guidelines in Oncology: Central Nervous System Cancers. Version 2.2026. Change as reported; wording to be confirmed against the guideline. nccn.org

MRI shown with all patient identifiers removed. MethyGlio results are from a manuscript under journal review and are data on file.