Somatic Variant Calling from UG 100 Data Using Deep Learning

The novel UG 100 sequencing platform enables cost effective analysis of somatic variants. Deep-learning based method was developed to call somatic variants on WGS and WES data and deep learning models were trained on mixtures of GIAB samples with different ratios to simulate somatic variants. We show that 100x coverage Tumor and 40x coverage of normal sample can be used to call variants down to VAF 10% in WGS. 500x tumor / 200x normal can be used to call variants down to VAF 5% in WES.