Machine learning accurately identifies the fragmentomic signature for DNase1L3 impairment caused by homozygosity for the p.Arg206Cys variant

DNase1L3 is an endonuclease that fragments DNA during apoptosis and digests DNA from microparticles in plasma, shaping key features of cell-free DNA (cfDNA) and maintaining extracellular DNA homeostasis, a process implicated in autoimmunity. We previously showed that homozygosity for the common DNASE1L3 p.Arg206Cys (R206C) variant is associated with a characteristic fragmentomic phenotype and reduced performance of fragmentomics-based cfDNA assays. Here, we investigated whether this phenotype itself can be recognized directly from cfDNA sequencing data using machine learning, thereby avoiding reliance on error-prone genotype imputation. To address this, we re-analyzed low-coverage whole-genome cfDNA sequencing data from 129,676 Non-Invasive Prenatal Tests (NIPT) and validated R206C genotypes in a selection of 169 matching plasma samples. Multiple machine-learning classifiers and unsupervised dimensionality reduction were applied to genome-wide fragmentomic features to identify the R206C-associated fragmentomic phenotype and to evaluate the minimum sequencing input required for accurate prediction. Our models outperformed genotype imputation for the identification of R206C homozygotes. Accurate predictions could be obtained with as few as 1,000 sequenced cfDNA fragments, substantially below the amount of sequencing routinely generated for cfDNA assays. Unsupervised analysis revealed few samples that cluster with homozygotes lack the corresponding genotype, suggesting that similar fragmentomic alterations may also arise through other mechanisms. Conversely, longitudinal follow-up across subsequent pregnancies showed some genetically confirmed R206C homozygotes initially lacked the aberrant fragmentome, but developed it in subsequent pregnancies. These findings show that machine learning enables the accurate identification of the R206C homozygote signature. It thereby provides a practical approach to improve cfDNA-based diagnostics across clinical applications and may serve as a biomarker for future DNase1L3-targeted therapeutic studies. The observed discordance between genotype and fragmentomic signature suggests that several features associated with DNase1L3 dysfunction are not fully explained by the genotype alone.