Table 1.

Summary of the cfDNA fragmentomics across different resolutions.

Methods	sequencing-based	Resolution	Reported/potential applications	Performance reported for cancer diagnosis	Tissue-of-origin inference	Potential limitations
Large-scale fragmentation patterns at megabase level (DELFI) [64]	Yes	5 Mb	Diagnosis of multiple early-stage cancers	57% to >99% sens@98% spec, AUC 0.94	Distinguish different cancer types (supervised way)	Performance is only evaluated by cross-validation. lack of a large independent prospective cohort for the validation. control is not age-matched. No application to the non-oncology field. low resolution limited the follow-up study to locate the potential therapeutic targets.
Large-scale co-fragmentation patterns (FREE-C) [65]	Yes	250 kb–1Mb	Cancer diagnosis (potential)	—	Absolute contribution value from different cell types	Lack of benchmark on the cancer diagnosis. low resolution limited the follow-up study to locate the potential therapetic targets. the reference panel for the tissues-of-origin estimation is arbitrary.
Fragment coverage near TSS [34]	Yes	±2 kb	Cancer diagnosis (potential)	—	—	Lack of benchmark on the cancer diagnosis. the gene expression prediction is only on a limited number of genes (housekeeping and always silenced) with binary prediction.
cfDNA-accessibility score near the transcription factor-binding sites (TFBS) [66]	Yes	±1 kb	Diagnosis of early-stage colon cancer	71% sens@72% spec for stage I, 74% sens@77% spec for stage II	Distinguish different cancer types (supervised way)	Performance is only evaluated by cross-validation. lack of a large independent prospective cohort for the validation. control is not age-matched. no application to the non-oncology field.
Orientation-aware cfDNA fragmentation (OCF) [67]	Yes	±1 kb	Diagnosis of early-stage HCC, organ transplantation, pregnancy	67.6% sens@ 93.8% spec	Estimate relative contributions from several cell types	Depending on the known open-chromatin regions. performance is only evaluated by cross-validation. lack of a large independent prospective cohort for the validation. control is not age-matched.
Windowed protection score (WPS) [33]	Yes	±120 bp	Cancer diagnosis (potential)	—	Estimation of several relevant cell types	Lack of benchmark on the cancer diagnosis. accurate nucleosome inference largely depends on the deep sequencing. many parameters in tissues-of-origin estimation are arbitrary
cfDNA-fragmentation hotspots [68]	Yes	200 bp	Diagnosis of multiple early-stage cancers	42% to 93% sens@100% spec	Distinguish different cancer types (supervised way)	Performance is only evaluated by cross-validation. lack of a large independent prospective cohort for the validation. control is not age-matched. no application to the non-oncology field.
Inference of DNA methylation from cfDNA fragmentation patterns [69]	Yes	Single base pair (high coverage) or 1 kb (low coverage)	Cancer diagnosis (potential)	—	Absolute contribution value from different cell types	Inference accuracy and resolution is the concern. lack of clinical applications in the study.
Preferred-ended position of cfDNA [70, 71]	Yes	Single base pair	Diagnosis of early-stage HCC, organ transplantation, pregnancy	AUC 0.88	Estimation of most relevant cell types	Accurate estimation of preferred-end sites requires deep sequencing. performance is only evaluated by cross-validation. lack of a large independent prospective cohort for the validation. control is not age-matched.
End-motif frequency and motif-diversity score (MDS) [72]	Yes	Global summary statistics	Diagnosis of multiple cancers, organ transplantation, pregnancy	AUC 0.89 (end motif), AUC 0.85 (MDS)	Estimation of most relevant cell types	MDS and end motif are summary statistics, their relationship with gene regulation at different location is not clear. performance is only evaluated by cross-validation. Lack of a large independent prospective cohort for the validation. Control is not age-matched.
Jagged end [73, 74]	Yes	Fragment level	Diagnosis of HCC, pregnancy	AUC 0.87 (JI-U in fragments 130–160 bp), AUC 0.54 (JI-U for all fragments)	Estimation of most relevant cell types	Performance is only evaluated by cross-validation. lack of a large independent prospective cohort for the validation. control is not age-matched.
Fragmentation patterns at eccDNA [76–78]	Yes	—	Cancer diagnosis (potential), pregnancy	—	—	Lack of benchmark on the cancer diagnosis and other studies.
LIQUORICE [131]	Yes	Integration of multiple resolutions	Diagnosis of Ewing sarcoma and other paediatric sarcomas	AUC up to 0.97	—	Fragmentation pattern in fine-scale depending on the known open-chromatin regions. cases are paediatric cancer, while healthy controls are all adults from other cohorts. lack of a large independent prospective cohort for the validation.
Enrich tumour signals by fragment size [60, 85]	Yes	Fragment level	Pan-cancer diagnosis	AUC 0.99 (late stage)	—	Control is not age-matched. sample size is small in each cancer category. cancer samples are late stage.
Filter CHIP-associated variants by fragment size [89, 90]	Yes	Single variant	Pan-cancer diagnosis (INVAR), early-stage lung cancer	AUC 0.98 (late stage, INVAR), 0.80 (early-stage, INVAR); 41% to 67%@98% specificity (Lung-CLiP, stage 1–3, risk matched control, independent prospective validation)	—	Classification is mostly based on genetic variants. control is not age-matched (INVAR). sample size is small in each cancer category (INVAR). Specialised sequencing technology (Lung-CLiP + CAPP-seq)
qPCR based	No	—	Cancers diagnosis, organ transplantation, pregnancy	—	—	Not genome-wide
Sophisticated capillary electrophoresis	No	—	Cancers diagnosis	—	—	Not genome-wide
Microscopy based	No	—	Cancers diagnosis	—	—	Not genome-wide

Non-NGS based methods are not discussed in detail.