Liquid-biopsy technology landscape

Leading cfDNA laboratories, platforms, and lessons for methylation research in colitis

More information from every cfDNA molecule

Modern liquid biopsy no longer asks only whether a mutation is present. Methylation, fragment size, fragment ends, nucleosome footprints, topology, and sequence variation can each carry information about disease and tissue of origin (Lo, Han, Jiang, and Chiu, 2021; Tsui, Jiang, and Lo, 2025). These technologies provide a roadmap for this project’s cfDNA work—but most were developed in cancer cohorts and cannot be assumed to perform similarly in ulcerative colitis.

Scientific illustration of methylation, fragmentation, immune, and statistical signals converging into a fingerprint

Multiple cfDNA features can be combined into an interpretable analytical fingerprint.

Leading laboratories

Institution / group Research leaders Distinctive contribution Relevance here
Johns Hopkins Kimmel Cancer Center Victor E. Velculescu; Robert B. Scharpf and collaborators DELFI uses genome-wide cfDNA fragmentation patterns; GEMINI models genome-wide mutation profiles from single cfDNA molecules. Their combination illustrates multi-feature detection. Fragmentation could complement methylation when inflammation, tissue injury, and neoplasia all contribute cfDNA.
CUHK Li Ka Shing Institute / Chemical Pathology Y. M. Dennis Lo; Rossa W. K. Chiu; Peiyong Jiang Foundational work in plasma DNA biology, fragment ends, topology, genome-wide methylation, and tissue-of-origin mapping. Directly informs the challenge of deciding whether a plasma methylation signal originates in colon, blood cells, or another tissue.
UCLA Zhou Lab Xianghong Jasmine Zhou; Wenyuan Li and collaborators CancerLocator, CancerDetector, cfTools, and cost-conscious cfDNA methylation analysis for detection and tissue localization. Offers computational precedents for read-level signal amplification and methylation-based deconvolution.
Stanford Oncology Maximilian Diehn; Ash Alizadeh; Billy Lau; Hanlee Ji and collaborators CAPP-Seq established ultrasensitive targeted ctDNA tracking; later native nanopore work jointly measures methylation and fragment length on individual cfDNA molecules. Shows a path from single-feature assays toward joint molecular profiles, although platform feasibility must be assessed locally.
Tagomics Jack Kennefick; John Gallon; academic and pharmaceutical collaborators Activace enriches informative unmethylated regions; LiquiPath uses cell-type methylation atlases and deconvolution for multi-organ safety monitoring. Demonstrates how tissue-specific reference atlases could make a blood signal more biologically interpretable.

Platform comparison

Platform Signal used Technical idea Current application Key caution
DELFI Fragment sizes and genome-wide fragmentation patterns Low-coverage whole-genome sequencing plus machine learning Cancer detection and treatment-monitoring research Fragmentation is affected by biology and pre-analytics; a cancer-trained classifier is not a UC classifier.
GEMINI Genome-wide mutation patterns in individual molecules Machine learning aggregates sparse mutation information Early cancer detection research, including combination with DELFI Somatic mutation burden and epigenetic dysregulation are different biological signals.
CancerLocator / CancerDetector cfDNA methylation Probabilistic tissue localization and joint CpG-pattern modeling at read level Research in early cancer detection and localization Reference methylomes, population transferability, and low tumor fraction determine performance.
Galleri Targeted cfDNA methylation Genomic sequencing, data science, and machine learning predict a cancer signal and likely origin Commercial multi-cancer early detection test; signals across more than 50 cancer types It complements rather than replaces recommended screening and does not detect every cancer. It is not validated for UC progression surveillance.
Native nanopore cfDNA Methylation plus physical fragment length on the same molecule PCR-free, single-molecule sequencing preserves linked epigenetic and fragmentomic features Emerging cancer-detection research Evidence is still developing; input, depth, error profiles, cost, and independent validation matter.
LiquiPath / Activace Cell-type-specific methylation and unmethylated-region enrichment Atlas-based deconvolution traces plasma cfDNA to contributing tissues Early-access drug-safety and organ-injury monitoring Proprietary platform; company-reported capabilities need independent, use-case-specific validation.

A practical technology ladder for this project

flowchart LR
  A["1 · Establish assay<br/>quality + pre-analytics"] --> B["2 · Compare tissue and<br/>plasma methylation"]
  B --> C["3 · Add tissue-of-origin<br/>deconvolution"]
  C --> D["4 · Evaluate fragment<br/>features"]
  D --> E["5 · Integrate methylation,<br/>sIgA + inflammation"]
  E --> F["6 · Validate in an<br/>independent cohort"]

  classDef base fill:#e9f5f2,stroke:#167d78,color:#123b5d;
  classDef advanced fill:#f7efe1,stroke:#c58b2c,color:#3c2d16;
  classDef validation fill:#123b5d,stroke:#123b5d,color:#fff;
  class A,B,C base;
  class D,E advanced;
  class F validation;

This sequence prevents technology from outrunning the biological question. With a modest exploratory sample, priority should go to reproducible sample handling, assay QC, pre-specified comparisons, and effect-size estimation. High-dimensional machine learning is most credible after sample size and external validation support it.

Design lessons that transfer

  1. Preserve provenance. Record collection tube, processing delay, centrifugation, storage, freeze–thaw history, DNA yield, library batch, and sequencing metrics.
  2. Model tissue of origin. Plasma is a mixture; methylation references and paired colon tissue can help separate epithelial biology from leukocyte or systemic signals.
  3. Keep features linked when possible. Single-molecule approaches can connect methylation and fragment length instead of measuring them in separate aliquots.
  4. Control the analytical search space. Small cohorts favor pre-specified regions, pathway-level summaries, shrinkage, and resampling over unconstrained classifiers.
  5. Separate discovery from validation. Internal cross-validation estimates model stability; it does not substitute for a new cohort, prospective follow-up, or clinical-utility testing.

Open the literature review → · Where this could fit in a clinic →

ImportantTranslation boundary

These platforms demonstrate what cfDNA can reveal in oncology and organ-injury settings. Their inclusion is a technology landscape, not evidence that they diagnose UC, predict colitis-associated CRC, or outperform colonoscopic surveillance.

Back to top

References

Y. M. Dennis Lo, Diana S. C. Han, Peiyong Jiang, and Rossa W. K. Chiu. (2021). Epigenetics, fragmentomics, and topology of cell-free DNA in liquid biopsies. Science. https://doi.org/10.1126/science.aaw3616.
W. H. Adrian Tsui, Peiyong Jiang, and Y. M. Dennis Lo. (2025). Cell-free DNA fragmentomics in cancer. Cancer Cell. https://doi.org/10.1016/j.ccell.2025.09.006.