Scalable Methods and Algorithms for Detecting Specific DNA Changes in Clinical Patients and Offspring of Radar-Exposed Personnel

Abstract

The abundance of Whole genome sequencing (WGS) data now enables research into more complex phenotypes and into effects of Ionizing radiation (IR) on human DNA. We used WGS data to contribute to four advances: 1. Finding a novel disease-causing variant for Koolen-de Vries syndrome; 2. Detecting mutational signatures in error-prone sequencing data; 3. Establishing a transgenerational biomarker for paternal exposure to IR; and 4. assessing Covid-19 host genetics. To ascertain potential transgenerational effects of IR exposure, we recruited a cohort comprised of 110 offspring of radar personnel of both German armies and accessed two more cohorts, one cohort of 130 offspring of liquidators and inhabitants of the town of Pripyat that were exposed to IR following the nuclear accident in 1986 (CRU cohort), and a large control cohort featuring 1214 offspring of non-exposed parents (Inova). We analyzed all data for the Radarstudy and Covid-19 data using newly developed WGS data analysis pipelines. Previous works suggested clustered de novo mutations (cDNMs), which are defined as two or more de novo mutations (DNMs) within 20 bp as potential signature of paternal IR exposure. To optimize the detection accuracy of DNMs and cDNMs, we used data from validation experiments to create a custom DNM and cDNM calling algorithm based on DeepTrio. We showed that deep-learning approaches based on DeepTrio can be trained with low data requirements. Our DNM detection model achieved a sensitivity of 95.7 % and a precision of 89.6 %. More complex mutational signatures, like cDNMs, can be detected with a precision of 76.9 % at 100 % sensitivity. Using newly developed analysis pipelines for WGS data, we detected a 4.7k bp deletion in <em>KANSL1</em>, that was found to be causing the patient phenotype, and provided insights into Covid-19 host genetics by elucidating correlations of variation and disease progression. When analyzing the three large WGS cohorts, we found that cDNMs were increased in children born to parents that were irradiated prior to conception. We observed 2.65 cDNMs per offspring on average in the CRUcohort, 1.48 in the Radar cohort and 0.88 in the Inova cohort (p < 0.005). Further statistical models indicated that this increase in cDNMs scales with the paternal exposure to IR (p < 0.001). These results leave little doubt that cDNMs represent a transgenerational biomarker of paternal IR exposure.