Method & Technology: cfDNA Methylation and Machine Learning

Galleri® is a multi-cancer early detection (MCED) screening test developed by GRAIL, Inc. that reads methylation "signatures" on circulating cell-free DNA (cfDNA). This article summarizes the molecular basis, workflow and technical limitations of the test.

Why Methylation?

Tumors leave distinctive methylation marks in the bloodstream even at the earliest stages of cancer. Methylation changes offer three advantages: they arise early, they produce strong signals across many detectable regions, and they carry tissue specificity so the signal can be traced to an anatomic region of origin. Galleri® targets these epigenetic signatures rather than mutation screening; this is how a single test can cover many cancer types.

Workflow

Sample: 2 Streck cfDNA tubes (venous blood) — no fasting required; can be drawn at home or at the center.
Transfer: the sample is prepared under Ministry of Health Biological Material Transfer System registration and chain-of-custody rules, and shipped by cold-chain courier to GRAIL's CLIA/CAP-accredited laboratory in the US.
Analysis: the cfDNA methylation profile is generated by next-generation sequencing; a machine-learning classifier computes the probability of a cancer signal.
Report: if no signal is found, "no cancer signal detected"; if a signal is found, the report includes the predicted tissue of origin (Cancer Signal Origin, CSO) among 18 anatomic regions.

Signal Score and CSO Prediction

The test output is a binary signal result (signal detected / not detected), and when a signal is found, a prediction of which of 18 anatomic regions it may originate from (Cancer Signal Origin, CSO). The CSO does not make a diagnosis; it helps the physician plan the diagnostic work-up (which imaging, which specialty).

Technical Notes and Limitations

The test is not an FDA-approved diagnostic; in the US it is a prescription-based laboratory-developed test (LDT) under CLIA.
Methylation-based screening can also produce abnormal signals in non-cancerous conditions (such as chronic inflammation or organ damage).
No screening test achieves 100% accuracy; results may only be evaluated by a physician.

References

Klein EA et al. Ann Intern Med. 2021;174(7):883-890 (CCGA — machine-learning classifier).
Schrag D et al. Ann Oncol. 2021;32(9):1167-1177 (PATHFINDER).
GRAIL Galleri® technical documentation (galleri.com).