rewirebio.iobenchmarks

Use caseDNA and genomesResearch and clinical research

Select a plasma ctDNA fragmentomics detection workflow

Which fragmentomics workflow detects tumour-derived plasma DNA under realistic low tumour fractions and fixed false-positive constraints?

What the evidence shows

Eight sources compare plasma cfDNA fragmentomic workflows on retrospective case-control cohorts.

None is a prospective screening study and none measures clinical benefit. Cristiano et al. 2019, the DELFI developers, report 73% sensitivity (152 of 208 cancers) at a reported 98% specificity for DELFI in repeated cross-validation against 215 healthy individuals. Hou et al. 2024 re-implemented ten published fragmentation patterns and DELFI with one support vector machine each on the same cohort.

They are independent of every pattern except IFS, which their corresponding author developed. In 10 x 10-fold cross-validation, pan-cancer AUC within open chromatin regions ranged from 0.8741 (fragment length) to 0.9736 (5' end motif), and their DELFI re-implementation reached 0.9575. Rankings changed on independent cohorts.

On 46 lung cancers and 385 healthy individuals, with a model trained on 12 lung cancers, the end-motif model fell to AUC 0.5355 and sensitivity 0.0652 at 95% specificity, while fragment size distribution reached 0.9386 and 0.7174. Hou et al.'s supplementary pan-cancer and liver cancer sensitivities conflict with other tables of the same file and are not shown. Wang et al. 2026, the UNITE developers, stratified cancers by ichorCNA tumour fraction in pooled public and new 0.1x data that include samples of the DELFI and Jiang et al. cohorts used above.

For cancers with an estimated tumour fraction of 3% or less, cross-validated AUC was 0.878 for all five features combined, 0.873 for fragment length, 0.857 for the SD of length counts, 0.736 for copy number, 0.715 for the C/T end-motif ratio and 0.651 for the short/long ratio, against 0.603 for a classifier using the ichorCNA tumour fraction alone, which detected 0.058 of these cancers at 95% specificity. For cancers with an estimated fraction above 3% and up to 10%, the ichorCNA classifier reached 0.983, above the combined model (0.962) and copy number alone (0.926). The extracted UNITE tables give sensitivity at a fixed specificity only for the ichorCNA comparator; the fragmentomic models' values are in a sheet not yet extracted.

Jamshidi et al. 2022 (Cancer Cell), by the GRAIL developers, scored fragment length and fragment endpoint classifiers from 30x cfDNA whole-genome sequencing on the same CCGA substudy 1 participants as their methylation, copy-number and variant classifiers, at a 98% specificity threshold set post hoc within each set. On the validation set of 464 cancers and 362 non-cancers the fragment length classifier detected 29% (95% CI 25% to 34%) and the fragment endpoint classifier 18% (15% to 22%), against 34% for whole-genome methylation analysed double-blinded; both fragment classifiers were developed after the validation set had been unblinded, so these are not blinded validation results, and no tumour-fraction-stratified sensitivity is printed. Zeng et al. 2026 (Nature Cancer), by the cfMeDIP-seq developers, computed fragmentomic features from cfMeDIP-seq reads, which are enriched for methylated fragments rather than drawn from whole-genome sequencing, and report means over 10 fold models.

On an independent paired-end validation set, a fragment-ratio classifier reached AUC 0.848 and detected 0.533 of cancers at 99% specificity, and fragment ratios combined with methylation reached 0.954 and 0.766. The validation set has 17 healthy controls, so its 99% specificity allows no false positive among them, and Li-Fraumeni syndrome carriers without current cancer count as cancers in its sensitivities. Three of the five classifiers that include end motifs scored below AUC 0.5 there, which the authors attribute to healthy controls processed with a different laboratory protocol, so validation results for all five are not shown.

Helzer et al. 2025 (Nature Communications) compared twelve fragmentomic metrics adapted to targeted exon panels, including the authors' own fragment-size entropy metrics, their combination and Griffin, with one classifier and shared folds, on plasma from patients with metastatic cancer. The healthy-versus-cancer one-vs-rest AUROC ranged from 0.835 (open-chromatin entropy) to 0.989 (first-exon depth) in a UW cohort of 404 cancers and 27 healthy donors, with Griffin at 0.944, and from 0.951 to 1.000 in public GRAIL panel data from 151 cancers and 47 healthy donors. In a developer comparison on 0.1x whole-genome sequencing (Cell Reports Medicine 2026), a fragment-level language model (Fragmentia-AI) reached AUC 0.9561 on a held-out set of unstated size, against 0.9209 to 0.935 for the developers' re-implementation of DELFI features and 0.881 to 0.8863 for Griffin features under three classifiers.

The same company's whole-genome model (Molecular Biomedicine 2026), on an independent test cohort of 2,952 cancers and 3,044 non-cancer controls, reached AUC 0.910 for cancers with an ichorCNA tumour fraction below 3% and detected 0.46 of them at a cutoff set to 98% specificity in training (observed test specificity 0.97); among those low-fraction cancers it detected 0.391 at stage I/II (637 of 1,628) and 0.708 at stage III/IV (345 of 487). Its tumour-fraction strata are ichorCNA estimates, no comparator was run on its cohorts, and its training cohort has the same number of non-cancer controls (1,245) as the Cell Reports Medicine cohort, so the two studies may share samples. Three of these four studies report on their authors' own method, and in the fourth the authors' own entropy metrics are among those compared.

Cohorts, feature implementations, classifiers and validation designs differ, so no value is comparable across sources.

You bring
  • Plasma cell-free DNA whole-genome sequencing reads
  • The target specificity (false-positive) constraint for the intended use
You want
Sourced AUC and sensitivity-at-fixed-specificity figures for fragmentomic workflows from internal cross-validation, independent validations and one developer same-participant comparison, and AUCs within ichorCNA tumour-fraction strata from one developer study; no prospective screening-validation claim.
Evidence in this release
9 comparisons shown · 43 tools · 88 configurations
Kept up to date
Evidence last checked 11 October 2026 · re-checked monthly, next due 11 November 2026
Executive Takeaway

For "Select a plasma ctDNA fragmentomics detection workflow", on Fragmentation patterns compared on the DELFI 2019 cohort (Hou et al. 2024, Table 1), AUROC (pan-cancer vs healthy; mean of 10 x 10-fold cross-validation) (AUROC, a proxy protocol), EDM scored 0.9824, reported by Systematically Evaluating Cell-Free DNA Fragmentation Patterns for Cancer Diagnosis and Enhanced Cancer Detection via Integrating Multiple Fragmentation Patterns.

Best reported valueEDMSystematically Evaluating Cell-Free DNA Fragmentation…
AUROC0.9824Fragmentation patterns compared on the DELFI 2019 cohort (Hou et al. 2024, Table 1), AUROC (pan-cancer vs healthy; mean of 10 x 10-fold cross-validation)
Baseline on this protocolNone scoredNo margin is computed
Ranked tools2523 from independent studies

25 tools have a value on this protocol; 23 of them come from studies whose authors did not develop the tool. No baseline method is scored on this protocol. Reviewers judged this protocol a proxy for the use case.

Ranking

Ranked on Fragmentation patterns compared on the DELFI 2019 cohort (Hou et al. 2024, Table 1), AUROC (pan-cancer vs healthy; mean of 10 x 10-fold cross-validation) (AUROC, higher is better). Proxy protocol Reviewers judged this protocol a proxy for the use case, not a direct test of it. Only results on this protocol are ranked; other tools are listed below the table.

11 tools in 25 rows, from 1 source. 23 of 25 results (92%) come from an independent study; 2 are reported by the tool's developers.

Rank▲ModelAUROC (higher is better)EvidenceCoverageReported byLicense
#1
EDMFragment 5' end motif (4-mer), region definition of the original publication (whole genome), SVM
0.9824Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#2
EDMFragment 5' end motif (4-mer), open chromatin regions after PCA dimension reduction, SVM
0.9737Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#3
EDMFragment 5' end motif (4-mer), open chromatin regions, SVM
0.9736Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#4
WPSWindowed protection score (WPS), open chromatin regions, SVM
0.9658Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#5
IFSIntegrated fragmentation score (IFS), open chromatin regions, SVM
0.9653Source-checked Developer-reported2/9Systematically Evaluating Cell-Free DNA…Not recorded
#6
WPSWindowed protection score (WPS), open chromatin regions after PCA dimension reduction, SVM
0.9649Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#7
IFSIntegrated fragmentation score (IFS), open chromatin regions after PCA dimension reduction, SVM
0.9641Source-checked Developer-reported2/9Systematically Evaluating Cell-Free DNA…Not recorded
#8
coverageFragment coverage (fragment midpoint counts), open chromatin regions after PCA dimension reduction, SVM
0.9640Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#9
endFragment endpoint counts, open chromatin regions, SVM
0.9639Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#10
coverageFragment coverage (fragment midpoint counts), open chromatin regions, SVM
0.9638Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#11
endFragment endpoint counts, open chromatin regions after PCA dimension reduction, SVM
0.9625Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#12
PFEPromoter fragmentation entropy (PFE), open chromatin regions, SVM
0.9579Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#13
PFEPromoter fragmentation entropy (PFE), open chromatin regions after PCA dimension reduction, SVM
0.9577Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#140.9575Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#15
OCFOrientation-aware cell-free fragmentation (OCF), open chromatin regions after PCA dimension reduction, SVM
0.9469Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#16
OCFOrientation-aware cell-free fragmentation (OCF), open chromatin regions, SVM
0.9467Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#17
FSRFragment size ratio (FSR), open chromatin regions after PCA dimension reduction, SVM
0.9447Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#18
FSRFragment size ratio (FSR), open chromatin regions, SVM
0.9441Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#19
FSDFragment size distribution (FSD), region definition of the original publication (chromosome arm), SVM
0.9431Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#20
FSDFragment size distribution (FSD), open chromatin regions, SVM
0.9271Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#21
FSDFragment size distribution (FSD), open chromatin regions after PCA dimension reduction, SVM
0.9229Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#22
lengthFragment length distribution, region definition of the original publication (whole genome), SVM
0.8835Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#23
lengthFragment length distribution, open chromatin regions after PCA dimension reduction, SVM
0.8763Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#24
lengthFragment length distribution, open chromatin regions, SVM
0.8741Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
#25
FSRFragment size ratio (FSR), region definition of the original publication (5mb bins), SVM
0.8456Source-checked Independent study2/9Systematically Evaluating Cell-Free DNA…Not recorded
32 tools are evaluated elsewhere in this use case but not on this protocol, so not ranked

Baseline

ichorCNA. ichorCNA estimates tumour fraction from genome-wide copy-number changes in shallow plasma sequencing, without training on cancer labels. Wang et al. 2026 use its tumour fraction, through a one-feature logistic regression, as a published comparator for their fragmentomic classifiers.

Machine-learning methods are compared with the baseline on the same data in one table: Fragmentomic feature sets by ichorCNA tumour fraction (Wang et al. 2026, UNITE). Baseline rows are marked in each table.

Tools compared

Comparisons

Unified model comparison for this use case. Compare tools against protocol baselines in a single view.

Fragmentation patterns compared on the DELFI 2019 cohort (Hou et al. 2024, Table 1)

Proxy evidenceIndependent study: Reported in a study whose authors did not develop the evaluated tool.Developer-reported: Reported by the developers of the evaluated tool.

Pan-cancer vs healthy ROC-AUC on the DELFI 2019 cohort for ten fragmentation patterns in open chromatin regions, with and without PCA, five patterns under their original region definitions and a DELFI re-implementation

Sorted by AUROC (higher is better). The best value in each column is highlighted.

ToolAUROC (pan-cancer vs healthy; mean of 10 x 10-fold cross-validation)
EDM3 sweepsFragment 5' end motif (4-mer), region definition of the original publication (whole genome), SVMmachine learning
0.9824 (best in column)
WPS2 sweepsWindowed protection score (WPS), open chromatin regions, SVMmachine learning
0.9658
IFS2 sweepsIntegrated fragmentation score (IFS), open chromatin regions, SVMmachine learning
0.9653
coverage2 sweepsFragment coverage (fragment midpoint counts), open chromatin regions after PCA dimension reduction, SVMmachine learning
0.9640
end2 sweepsFragment endpoint counts, open chromatin regions, SVMmachine learning
0.9639
PFE2 sweepsPromoter fragmentation entropy (PFE), open chromatin regions, SVMmachine learning
0.9579
DELFImachine learning
0.9575
OCF2 sweepsOrientation-aware cell-free fragmentation (OCF), open chromatin regions after PCA dimension reduction, SVMmachine learning
0.9469
FSR3 sweepsFragment size ratio (FSR), open chromatin regions after PCA dimension reduction, SVMmachine learning
0.9447
FSD3 sweepsFragment size distribution (FSD), region definition of the original publication (chromosome arm), SVMmachine learning
0.9431
length3 sweepsFragment length distribution, region definition of the original publication (whole genome), SVMmachine learning
0.8835
Caveats and method
  • Internal 10 x 10-fold cross-validation on one cohort
  • AUC only; no sensitivity at a fixed specificity
  • IFS was defined by Hou et al.'s corresponding author (Zhou et al. 2022), so its evaluations are author-reported; the other patterns and DELFI are independent re-implementations by Hou et al., not the original tools.
  • Same cohort as the DELFI judgement (sensitivity at 98% specificity); the UNITE cross-validation set also draws 352 samples from the Cristiano et al. 2019 FinaleDB data.

Protocol: Hou et al. 2024 pan-cancer vs healthy, Cristiano cohort, 10 x 10-fold cross-validation (Table 1 AUC)

Primary benchmark evaluation shown (Fragmentation patterns compared on the DELFI 2019 cohort (Hou et al. 2024, Table 1)). 8 additional study protocols (Fragmentation patterns at fixed specificity across cohorts (Hou et al. 2024, Tables S2-S3), Fragmentomic feature sets by ichorCNA tumour fraction (Wang et al. 2026, UNITE), CCGA substudy 1: cfDNA classifiers on the same participants at a post hoc 98% specificity and others) are cited in individual tool drawer rows and under Question, sources and review.

Not covered yet

  • 4 of 215 healthy individuals were misclassified at the source-labelled 98% specificity; the printed specificity is retained rather than recomputed.
  • cfDNA signal reflects total circulating DNA, not purified ctDNA.
  • Tumour-fraction-stratified evidence comes from two developer studies: cross-validation in Wang et al. 2026 and an independent test cohort in the Fragmentia-AI WGS paper (Molecular Biomedicine 2026), both using ichorCNA estimates rather than measured tumour fractions. In the extracted UNITE tables, sensitivity at a fixed specificity within a tumour-fraction stratum is printed only for the ichorCNA tumour-fraction comparator. The fragmentomic models' values are in Data file S2 sheet STATS_xgb_all_feat, which was not extracted; that sheet prints identical sensitivity at 98% and 99% specificity for every feature set in the lowest stratum and needs checking before use. The Fragmentia-AI WGS strata cover one model with no comparator.
  • Hou et al. 2024 Supporting Information Table S2 pan-cancer and liver cancer sensitivities conflict with other tables of the same file and are not shown.
  • No independent dilution-series or limit-of-detection comparison of fragmentomic tools (Griffin, LIQUORICE, FinaleToolkit, DELFI) with printed values was found.
  • No specificity in patients with inflammatory, autoimmune or vascular disease was ingested; Curtis et al. 2025 (PNAS) report false positives in these groups with per-sample data only.
  • No result is comparable across sources: cohorts, feature implementations, classifiers and validation designs differ, and the sources share samples of the DELFI 2019 and Jiang et al. cohorts.
  • Jamshidi et al. 2022 fragment endpoint, fragment length and allelic imbalance classifiers were developed after validation blinding was lifted, so their validation values are not a blinded validation; their clinical limits of detection by tumour allele fraction are in figures only.
  • Helzer et al. 2025 compare panel fragmentomic metrics at low ctDNA fraction by in silico mixing, but the values are in Figure 6B only.
  • The Fragmentia-AI (Cell Reports Medicine 2026) and Fragmentia-AI WGS (Molecular Biomedicine 2026) cohorts both come from the Geneseeq repository and report the same number of non-cancer controls (1,245) at 0.1x, so they may share samples; the Fragmentia-AI WGS external cohort is the public DELFI cohort, which overlaps the DELFI 2019 samples above.
  • Zeng et al. 2026 compute fragmentomic features from cfMeDIP-seq (methylation-enriched) reads, not whole-genome sequencing, and Helzer et al. 2025 from targeted exon panels.

Contribute evidence or propose a correction

How to cite

Select a plasma ctDNA fragmentomics detection workflow. Rewire biological model benchmark database, release 2026-11-21-c8bea25e5abb. https://benchmarks.rewirebio.io/use-cases/plasma-ctdna-fragmentomics/

Release 2026-11-21-c8bea25e5abb. Rankings and values change between releases, so keep the release id in the citation. Cite the original publication as well for any value you reuse: each result links its source.

The compilation and annotations made by Rewire in this release are offered under CC BY 4.0; values and text taken from cited sources stay under the terms of those sources and are not covered by this licence. Data licence

Question, sources and review

Who this is for, setting and exclusions

Compare the fragmentomic feature sets and classifiers with printed detection results on shared plasma cohorts, including one source stratified by ichorCNA tumour fraction, before selecting a workflow and a validation design with a fixed specificity for the intended screening or diagnostic population.

Who this is for
  • Clinical researchers scoping the available diagnostic-genomics evidence
  • Computational researchers comparing exact evaluated configurations
Biological setting
Retrospective case-control plasma cfDNA whole-genome sequencing. The DELFI classifier achieves 73% (152 of 208) sensitivity at a reported 98% specificity in repeated 10-fold cross-validation (10 repeats) across 208 cancer patients (seven cancer types) and 215 healthy individuals. Hou et al. 2024 re-implemented ten published fragmentation patterns and DELFI with one support vector machine each on the same cohort, on a liver cancer cohort and on three independent validation cohorts. Wang et al. 2026 (UNITE) cross-validated five shallow-WGS fragmentomic feature sets, their combination and an ichorCNA tumour-fraction classifier within ichorCNA tumour-fraction strata, on pooled public and new data that include samples of the DELFI and Jiang et al. cohorts. Jamshidi et al. 2022 (CCGA substudy 1) scored fragment endpoint and fragment length classifiers from 30x cfDNA whole-genome sequencing against methylation, copy-number and mutation classifiers on the same participants at a post hoc 98% specificity.
Clinical research scope
Clinical applicability is not established. Every result comes from a retrospective case-control cohort, with internal cross-validation, small independent validations (8 to 129 cancers) or one developer validation set (464 cancers) on which the fragment classifiers were developed after unblinding, not from a prospective screening trial. The clinically identified cancers and healthy comparators differ from an intended screening population, and tumour fractions are ichorCNA estimates, not measured values.

Outside this use case

  • A combined mutation+DELFI configuration (115/126, 91%), a different subset/configuration not ingested here
  • Prospective screening performance and clinical benefit: no prospective or intended-use cohort was ingested
Evidence collection plan
Evidence collection planCollecting evidence

Already on record: 9 comparisons shown · 43 tools, in the evaluated evidence above. The status above describes only the specific comparison in this plan, which remains open; it does not mean no evidence has been collected. The plan defines a comparison to investigate; it does not establish model performance or suitability.

Comparison question

Which fragmentomics workflow detects tumour-derived plasma DNA under realistic low tumour fractions and fixed false-positive constraints?

Baselines, outcomes and validation requirements

Baselines to include

  • The conventional/author-introduced workflow measured in the linked primary source(s).

Outcomes to measure

  • The declared endpoint in this use case's active mapping(s); see evidence_gaps for what remains open.

Validation requirements

  • Independent held-out population matched to the intended clinical setting.
  • Qualified human scientific review before any clinical-validation claim.

Next collection task

A dedicated rewire-benchmarks protocol/run task with an independent, prospectively collected held-out validation cohort.

Sources, review and downloads

Automated source review · 2026-10-07 · Claude Sonnet AMP-integration worker, bounded transcription of Codex-checked primary values; independently reviewed by Codex (workbench/amp-supervision/primary-review.md, integration-review-corrections.md)

Bounded primary-source transcription, independently Codex-checked. No new model execution, independent experimental reproduction, qualified human scientific review or clinical validation.

Release 2026-11-21-c8bea25e5abb · question use-case-plasma-ctdna-fragmentomics

Use-case input digest bfbb7ab9303becad1e3cb395568c26f41e72ddb67cd1a89a06d9eaa0f76bd95f

Download questions and review metadata (JSON) (gzip) · Verify release checksums (gzip)