O4 Using somatic data to aid germline clinical variant interpretation in developmental disorders

Obtaining a molecular diagnosis for children with developmental disorders (DD) is transformative for their medical care, but variants of uncertain significance are common. We investigated whether using somatic mutation data profiling the driver mutation landscape in cancer and healthy human tissues could improve germline variant interpretation in DD. A logistic regression model was used to combine a computational effect predictor with somatic data from cancer (COSMIC) and healthy tissues (sperm and buccal Nanoseq data described previously). The model was trained to distinguish germline pathogenic variants (Clinvar) from likely benign variants (UK Biobank). This combined model had a significantly greater area under the receiver operating characteristic curve (AU-ROC) in held out test data than a computational effect predictor alone in certain gene groups. The greatest diagnostic uplift was in altered function mechanism DD genes that are also oncogenes, particularly the RASopathy genes (combined model AU-ROC 0.963 [95% CI 0.956-0.970], AlphaMissense AU-ROC 0.912 [0.905-0.919]). 24 DD genes were identified with a significantly better AU-ROC when somatic data is added (bonferroni corrected DeLong test p value < 0.05). Increasing the power could potentially identify many more genes.