Genetics

Pan-cancer cfDNA methylome and fragmentome atlas enables robust multicancer detection

This study aggregates 1,074 cfMeDIP-seq plasma profiles from nine studies, spanning 11 cancer types, Li-Fraumeni carriers, and healthy controls. A harmonized pipeline identifies 14,202 pan-cancer differentially methylated regions for detection plus cancer-specific markers for subtype monitoring. Fragmentomic features, including 5' end motifs, fragment length distributions, and nucleosome footprints, also differ systematically across cancers. Integrating methylation and fragmentomic features enhances cancer detection and classification relative to either modality alone. Validation in 220 independent samples, including three cancer types absent from discovery, supports robustness across indications and cohorts. 1

M-PACT accurately classifies pediatric CNS tumors from subnanogram CSF cfDNA methylomes

M-PACT is a deep neural network that classifies CNS tumors using methylation profiles from subnanogram-input CSF cfDNA. In an embryonal CNS tumor benchmarking cohort of 79 cases, M-PACT achieved 92% classification accuracy. In an independent validation cohort of 58 embryonal cases, accuracy remained high at 88%. The workflow also enables methylation-based cellular deconvolution and sensitive copy-number variation detection from low-input cfDNA. Performance in nonembryonal tumors, balanced genomes, and nonmalignant CSF suggests broad diagnostic applicability, warranting prospective clinical trials. 2

Genome-first analysis quantifies inherited pathogenic variant burden across diverse populations

Using gnomAD, Regeneron Million Exome, and Turkish Variome data, the authors interrogated 4,591 PanelApp and OMIM disease genes. An ACMG-based classifier identified 97,135 pathogenic and 478,263 likely pathogenic variants, expanding the catalog nearly six-fold. Individuals carry on average 4.70 pathogenic or likely pathogenic variants, with 1.66 compatible with a Mendelian condition at the genotype level. About 1 in 11 people has an actionable genotype, and 382 genes emerge as candidates for carrier screening. A genome-first framework estimates genotype-compatible disease across 13 ICD-10 groups, highlighting opportunities for preventive genomics and risk counseling. 3

Aggregated DDX41 datasets refine gene-specific variant classification in heritable MDS and AML

This analysis compiles a synthetic cohort of germline and somatic DDX41 variants from 35 studies, plus 832 additional reported cases. Deleterious germline DDX41 variants are confirmed as the leading cause of heritable predisposition to myelodysplastic neoplasia and acute myeloid leukemia. Refined ACMG and AMP criteria leverage case enrichment (PS4), germline–somatic associations (PP4), and improved computational prediction (PP3 and BP4). A Bayesian multinomial model uses somatic patterns to update odds of pathogenicity, improving classification of germline variants. AlphaMissense outperforms REVEL in sensitivity for DDX41 missense prediction, and results are implemented in an online curation tool for consistent practice. 4