Whole-exome sequencing capture kit biases yield false negative mutation calls in TCGA cohorts

Abstract

The Cancer Genome Atlas (TCGA) provides a genetic characterization of more than ten thousand tumors, enabling the discovery of novel driver mutations, molecular subtypes, and enticing drug targets across many histologies. Here we investigated why some mutations are common in particular cancer types but absent in others. As an example, we observed that the gene CCDC168 has no mutations in the stomach adenocarcinoma (STAD) cohort despite its common presence in other tumor types. Surprisingly, we found that the lack of called mutations was due to a systematic insufficiency in the number of sequencing reads in the STAD and other cohorts, as opposed to differential driver biology. Using strict filtering criteria, we found similar behavior in four other genes across TCGA cohorts, with each gene exhibiting systematic sequencing depth issues affecting the ability to call mutations. We identified the culprit as the choice of exome capture kit, as kit choice was highly associated with the set of genes that have insufficient reads to call a mutation. Overall, we found that thousands of samples across all cohorts are subject to some capture kit problems. For example, for the 6353 samples using the Broad Institute’s Custom capture kit there are undercalling biases for at least 4833 genes. False negative mutation calls at these genes may obscure biological similarities between tumor types and other important cancer driver effects in TCGA datasets.

A verified genomic reference sample for assessing performance of cancer panels detecting small variants of low allele frequency

“…We investigated the potential amount of bias in individual VAF estimates that may be associated with each kit and may be a unique characteristic with certain variants. Bias in VAF can arise from several factors: bait-hybridization bias, mapping errors and bias in mapping towards the reference genome [ 36 ], and confusion in calling a multiple nucleotide polymorphism or a complex multi-allelic polymorphism. To separate mapping and calling issues from bait biases, we ran several individual pipelines including a compendium method (SomaticSeq) for each kit of pooled Sample A replicates as well as in silico Sample A and merged-BAM Sample A (see “ Methods ” for details).…”

Impact of variant-level batch effects on identification of genetic risk factors in large sequencing studies

“…If not adequately addressed in the analysis, batch effects reduce statistical power and lead to both false-positive and false-negative associations. Practices in large sequencing studies that commonly introduce batch effects include dividing samples among multiple sequencing centers [2,3], collecting or preparing samples under different protocols [4], and extracting exomes using different target capture kits [4,5]. For example, the Alzheimer's Disease Sequencing Project (ADSP) sequenced exomes of more than 10,000 cases and controls to identify genetic factors associated with Alzheimer's disease (AD) [6].…”

Methodological differences can affect sequencing depth with a possible impact on the accuracy of genetic diagnosis

“…It is well known that the efficiency of capture depends on several experimental procedures as well as on probe design, which may directly affect sequence depth and uniformity (Do et al, 2012;Chandler et al, 2016). Therefore, problems in the capturing reaction directly affect the final experiment results, yielding not only regions with different average depths but also leading to regions with no coverage at all (Lionel et al, 2018;Wang et al, 2018). We demonstrated here that differences, most likely attributed to the different methods used by the sequencing centers, proved to play a significant role in determining the distribution of sequencing depth in WES data from the 1000 Genomes Project.…”