OncoFirm™ / Research / Liquid biopsy

Research explainer · Liquid biopsy

Early Cancer Detection Using Cell-Free DNA (ccfDNA): What the MIGHT Studies Mean for Diagnostics

Blood tests that read cell-free DNA are one of the most active areas in early cancer detection. Two 2025 PNAS studies from Johns Hopkins researchers show both the promise and a hard limit: a new AI method, MIGHT, made cfDNA results more reliable, while a companion study found that inflammation can produce cancer-like DNA signals. Here is what the research found, what it did not show, and what it means for multi-analyte diagnostics.

By OncoFirm™ Scientific TeamPublished 27 Aug 2025Updated 1 Oct 2026Sources peer-reviewed, linked

At a glance

Topic

Circulating cell-free DNA (ccfDNA) for cancer detection

Key studies

Curtis et al., PNAS 2025 (MIGHT) and companion PNAS 2025 study on fragmentation signatures

Headline result

72% of advanced cancers detected at 98% specificity (primary analysis)

Main caution

Autoimmune and vascular inflammation can mimic cancer signals

Disclosure

OncoFirm™ was not involved in these studies

On this page

Key findings

The research in brief

72%of advanced cancers detected at 98% specificity, using aneuploidy features (102 cancers, 250 controls)
44sets of cfDNA features compared to find which signals were most reliable
Inflammationproduced cancer-like fragmentation patterns in autoimmune and vascular disease

Figures are from the PNAS papers and the Johns Hopkins news release; see the references for full details.

Background

What is ccfDNA, and why use it for cancer detection?

Circulating cell-free DNA (ccfDNA) is short DNA fragments released into the blood as cells die. In people with cancer, a small fraction comes from the tumor and can carry tumor-specific changes, including mutations, abnormal chromosome numbers (aneuploidy) and distinctive fragment lengths and end patterns, known as fragmentomics.

Because one blood draw can sample DNA from many tissues, ccfDNA tests are being studied for multi-cancer early detection. The challenge is that tumor DNA is often a tiny fraction of the total, especially in early disease. For background see liquid biopsy explained and circulating tumor DNA (ctDNA).

Two liquid-biopsy signals from one blood sampleA blood sample is separated into plasma. One path sequences cell-free DNA fragments to measure fragmentation and aneuploidy features, scored by an uncertainty-aware model such as MIGHT. The other path measures protein antigens such as CEA by immunoassay, including fluorescent point-of-care tests. Both can feed multi-analyte interpretation, with inflammation as a known confounder.Blood sampleplasma separatedCELL-FREE DNAFragmentslength, ends, positionsSequencingcentral laboratoryFeaturesaneuploidy & moreModel scorecalibrated (MIGHT)PROTEIN ANTIGENSAntigense.g. CEA, CA-125Immunoassaylab or point of careMeasured valueng/mL, traceableTrendserial resultsMulti-analyteINTERPRETATIONKnown confounder: inflammation (autoimmune and vascular disease) can mimic cancer signals.

One blood sample, two kinds of signal. DNA-based tests read fragments by sequencing in a central laboratory; protein tests measure antigens such as CEA by immunoassay, including fluorescent point-of-care formats. Combining them, and accounting for inflammation, is an active research direction.

The MIGHT method

How MIGHT aims to make AI decisions more reliable

Medical datasets often have thousands of measured variables but only hundreds of patients. Standard machine-learning models can look accurate in development and then fail to reproduce. MIGHT (Multidimensional Informed Generalized Hypothesis Testing) combines ensembles of decision trees with careful calibration and cross-validation so that its decisions come with quantified uncertainty, and so it can be tuned to a required specificity.

  • Feature comparison: the team evaluated 44 different sets of cfDNA features, including fragment lengths, positions, end motifs and aneuploidy measures.
  • Best performer: aneuploidy-based features, reflecting abnormal chromosome numbers in cancer cells.
  • Result: 72% sensitivity at 98% specificity in the primary analysis of advanced cancers.
  • CoMIGHT: an extension that combines several variable sets, tested in cohorts including early-stage breast and pancreatic cancers.
  • Open science: MIGHT and CoMIGHT were released for research use through the treeple project.

The inflammation problem

Why inflammation can look like cancer in cfDNA

The companion study found that fragmentation signatures thought to be characteristic of cancer also appeared in some patients with autoimmune diseases, including lupus, systemic sclerosis and dermatomyositis, and with vascular conditions such as venous thromboembolism. The shared patterns tracked with markers of inflammation.

When the researchers added samples from these patients to MIGHT’s training data, the model got better at telling cancer apart from inflammatory disease. False positives fell but were not eliminated. The broader lesson for every cancer biomarker is the same: specificity has to be tested against the conditions most likely to imitate the target, not only against healthy volunteers.

Limitations

What these studies did not show

  • The headline 72% figure comes from patients with advanced cancers, chosen to maximize signal; early-stage performance is typically lower.
  • These are research studies, not validated clinical tests, and they were not prospective screening trials.
  • Results depend on sequencing depth, cohort composition and sample handling, and need independent replication.
  • A correction to the companion study was published in 2026; readers should consult the current version.

Where protein markers fit

DNA and protein: complementary signals

Protein tumor markers such as CEA and CA-125 have been measured in blood for decades. They are not specific enough to screen populations on their own, but they are inexpensive, quantitative and well suited to monitoring. Earlier Johns Hopkins work on the multi-analyte CancerSEEK test combined cfDNA mutations with eight proteins, including CEA and CA-125, because the two kinds of signal catch different tumors.

Protein immunoassays also have their own confounders, including inflammation, smoking and benign disease, so the MIGHT studies’ emphasis on calibrated, uncertainty-aware decisions applies equally to them. More in tumor antigen detection and AI multi-biomarker blood tests.

What it means for OncoFirm™

Lessons we are building into our platform

Quantitative protein measurement

A self-calibrating fluorescent strip and handheld reader, starting with quantitative CEA and PSA. Platform

Specificity by design

Validation plans that test against inflammatory and benign conditions, not only healthy controls, following the lesson of the companion study.

Uncertainty-aware software

Locked, validated models that report quality flags and confidence with every result. AI-assisted diagnostics

Multi-analyte research

Collaborations that pair protein panels with other data types, including cfDNA, in research settings. Clinical collaborations

See how point-of-care protein testing compares with laboratory testing in fluorescent LFA vs. traditional testing.

FAQ

Frequently asked questions

What is ccfDNA?

Circulating cell-free DNA (ccfDNA, often written cfDNA) is short fragments of DNA released into the bloodstream as cells die. In people with cancer, a small share comes from tumor cells and can carry tumor-specific changes in sequence, chromosome number and fragment patterns.

What is MIGHT?

MIGHT (Multidimensional Informed Generalized Hypothesis Testing) is a machine-learning method from Johns Hopkins researchers, published in PNAS in 2025, designed to give reliable, calibrated decisions with quantified uncertainty when there are many variables and relatively few samples.

How accurate was MIGHT at detecting cancer from ccfDNA?

In the study's primary analysis of 102 people with advanced cancers and 250 without cancer, the best aneuploidy-based feature set detected 72% of cancers while keeping specificity at 98%. Early-stage performance was not the focus of that analysis.

Why can inflammation cause false positives in cfDNA tests?

A companion PNAS study found that cfDNA fragmentation patterns seen in cancer also appear in some patients with autoimmune diseases such as lupus and in vascular conditions such as venous thromboembolism, and that these patterns tracked with markers of inflammation.

Is MIGHT a clinical test?

No. MIGHT is a research method. The researchers made MIGHT and CoMIGHT openly available for research use, but neither is an approved diagnostic test.

Is OncoFirm involved in the MIGHT research?

No. OncoFirm was not involved in these studies. This article summarizes published, peer-reviewed research and explains how its lessons apply to protein-based diagnostics that OncoFirm is developing.

Sources

References

  1. Curtis SD, Panda S, Li A, et al. Minimizing and quantifying uncertainty in AI-informed decisions: Applications in medicine. PNAS. 2025;122(34):e2424203122. doi.org/10.1073/pnas.2424203122
  2. Curtis SD, et al. Fragmentation signatures in cancer patients resemble those of patients with vascular or autoimmune diseases. PNAS. 2025;122:e2426890122. A correction was published in 2026 (doi:10.1073/pnas.2538063123). doi.org/10.1073/pnas.2426890122
  3. Johns Hopkins Medicine. New method advances reliability of AI with applications in medical diagnostics. News release, August 2025. www.hopkinsmedicine.org/news/newsroom/news-releases/2025/08/new-method-advances-reliability-of-ai-with-applications-in-medical-diagnostics
  4. Cohen JD, Li L, Wang Y, et al. Detection and localization of surgically resectable cancers with a multi-analyte blood test. Science. 2018;359:926–930. doi.org/10.1126/science.aar3247
  5. National Cancer Institute. Tumor Markers. www.cancer.gov/about-cancer/diagnosis-staging/diagnosis/tumor-markers-fact-sheet

About this article. Written by the OncoFirm™ Scientific Team from peer-reviewed publications, which are linked above. OncoFirm™ was not involved in the studies discussed and has no affiliation with their authors. This article is for education and is not medical advice. OncoFirm™ assays are in development, have not been cleared or approved by the FDA and are not available for sale.

Collaborate on multi-analyte cancer detection

Research groups working with cfDNA, protein biomarkers or AI can explore joint studies with OncoFirm™. Collaboration inquiry

[email protected]+1 516.900.2606Port Washington, NY