Methylation sequencing also carries fragmentomics: plasma cfDNA blocks give cancer tests an interpretable second signal

Methylation-sequencing libraries can serve as a source of fragmentomic information, not only as carriers of methylation reads. Fragmentomic features were usually derived from whole-genome sequencing, and whether methylation workflows preserve them had not been systematically assessed. In healthy and gastric cancer plasma cohorts, whole-genome bisulfite sequencing and enzymatic methyl sequencing both recapitulated the fragment length distributions and nucleosome-associated patterns near transcription start sites seen in whole-genome sequencing, though cross-platform concordance of some genome-wide features varied by cohort.

Plasma cfDNA itself can be segmented into co-methylated blocks that serve as shared intervals for multimodal signals. Earlier block-based assays typically defined blocks from tissue methylomes and then quantified them in plasma. From plasma enzymatic methyl sequencing of seven cancer types and healthy controls, 151,473 blocks were identified with median lengths of 30–50 bp, shorter than tissue-derived blocks; re-identification in an independent validation cohort matched 86.0% of the original blocks.

Methylation and fragmentomic signals are coordinated yet complementary within the same blocks, and the resulting ensemble model is competitive across several cancer tasks. Single-modality models used only one class of methylation or fragmentomic information. In a breast cancer independent validation cohort the genome-wide-augmented model achieved the highest AUC; in an external hepatocellular carcinoma cohort it ranked among the top methods, while SPOT-MAS achieved the highest AUC in that small cohort.

The report covers how well fragmentomic signals survive whole-genome methylation sequencing of plasma cfDNA, and how plasma-derived co-methylated blocks work as interpretable units for multimodal organization; model evaluation spans cross-validation, independent validation and two external cohorts across seven cancer types and healthy controls. Using this framework for screening or monitoring would require larger prospective studies with more early-stage cancers, broader benign-disease controls and longitudinal samples.

Cross-platform comparisons used biologically comparable cohorts rather than the same plasma aliquots, so protocol effects cannot be fully separated from inter-cohort biological variation. Comparisons with SPOT-MAS, THEMIS and DELFI followed published descriptions as closely as possible, but differences in preprocessing and implementation details mean performance differences should be read in that context. Red and blue blocks are explicitly operational labels rather than universal chromatin-state annotations.