Improving somatic exome sequencing performance by biological replicates

This study aims to enhance somatic variant calling performance and develop ML models using different replicate-based consensus approaches. We examine the effect of combining multiple biological replicates on variant calling performance in the WES datasets from the SEQC2 consortium. We conduct comparisons to evaluate the potential improvements in variant calling performance by using replicates sourced from the same center as well as those obtained from different centers. We also train ML models using these replicates and achieve performance comparable to optimal models trained using declared high-confidence variants.