Abstract 2591: A guide to filtering TARGET Complete Genomics germline variants

Abstract

Abstract With the public release of TARGET’s pediatric cancer data, researchers are eager to investigate and integrate variants from both tumor and germline sequencing data, but caution should be used when interpreting TARGET variant calls from Complete Genomics data, as they unlike the Illumina based calls for which most of our databases and tools are customized. Here, we use germline samples from 55 neuroblastoma cases with both whole genome Complete Genomics variants and whole exome Illumina variants from GATK, Pindel, Platypus, and bam2mpg to investigate differences between germline variant calls from Complete Genomics and Illumina platforms. Our study of ~160,000 single nucleotide polymorphisms (SNPS) and ~19,000 insertions or deletions (indels) reveals that while SNP sets largely agree between Complete Genomics and Illumina, Complete Genomics data have many indels not present in Illumina calls, making it difficult to screen Complete Genomics indels against popular Illumina based variant databases like those provided by the Exome Aggregation Consortium and the National Heart, Lung, and Blood Institute Exome Sequencing Project. Further probing of Complete Genomics indels uncovered inconsistent indel formatting resulting a loss of recurrent variants, and miscalculated read depths in earlier versions of the Complete Genomics platform. To solve these issues and make Complete Genomics variant calls comparable to Illumina calls, we introduce a Snakemake pipeline that reformats Complete Genomics variant files and creates variant features for use in decision tree or random forest classifiers trained to recognize Complete Genomics variants that will be present in Illumina variant call sets. We determine that the most important features for locating Complete Genomics variants likely to be found in Illumina germline variant call sets are the presence of reads from both directions, average variant allele fraction across the cohort, a variant call in the matched tumor, the variant fraction of reads, and read depth. Using ten fold cross validation with our decision tree, we obtain >95% recall and >98% precision for SNPs, and >90% recall and >85% precision for indels. We advocate the use of our pipeline and filtering scheme for making Complete Genomics variant calls comparable to Illumina based calls. Citation Format: James P. Evans, Rajesh Patidar, Zalman Vaksman, Sivasish Sindiri, Douglas R. Stewart, Javed Khan, Jun S. Wei, Sharon J. Diskin. A guide to filtering TARGET Complete Genomics germline variants [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 2591. doi:10.1158/1538-7445.AM2017-2591