A database of simulated tumor genomes towards accurate detection of somatic small variants in cancer

Abstract

AbstractBackgroundSomatic mutations promote the transformation of normal cells to cancer. Accurate identification of such mutations facilitates cancer diagnosis and treatment. A number of callers have been developed to predict them from paired tumor/normal or unpaired tumor sequencing data. However, the small size of currently available experimentally validated somatic sites limits evaluation and then improvement of callers. Fortunately, NIST reference material NA12878 genome has been well-characterized with publicly available high-confidence genotype calls.ResultsWe used BAMSurgeon to create simulated tumors by introducing somatic small variants (SNVs and small indels) into homozygous reference or wildtype sites of NA12878. We generated 135 simulated tumors from 5 pre-tumors/normals. These simulated tumors vary in sequencing and subsequent mapping error profiles, read length, the number of sub-clones, the VAF, the mutation frequency across the genome and the genomic context. Furthermore, these pure tumor/normal pairs can be mixed at desired ratios within each pair to simulate sample contamination.ConclusionsThis database (a total size of 15 terabytes) will be of great use to benchmark somatic small variant callers and guide their improvement.Contact informationjing.mengrabbit@gmail.com