Abstracts

Rationale: Whole-exome sequencing (WES) is an effective method for the detection of disease-causing variants. Although the computational pipelines for detection of single-nucleotide variants or small insertions/deletions have been well established, identifying copy number variation (CNV) from WES data remains a challenge. Many tools have been developed for CNV detection based on WES. However, these programs usually show discordant results among the algorithms. In this study, we estimated the performance of programs and considered whether selecting WES as the first-tier clinical test was appropriate for CNV detection.

Methods: WES data of 180 individuals with neurologic disorders including epilepsy and developmental delay were produced using the Agilent SureSelect V6 target capture kit. Among these samples, 44 individuals also had chromosomal microarray analysis (CMA) data. CMA results were considered as a reference set and compared with the results of CNV analysis from WES to estimate the sensitivity and positive predictive value (PPV) of the CNV detection. We selected four CNV detection tools (CNVkit, CoNIFER, ExomeDepth, and cn.MOPS) and gauged the performance of the programs.

Results: In this work, we found that the sensitivity was highest in ExomeDepth, followed by CoNIFER, CNVkit, and cn.MOPS (74.26%, 55.45%, 38.61%, and 19.80%, respectively). The mean size of CNVs identified in all CNV callers was 1.2 Mb, and the mean CNV sizes detected in three, two, and the single caller were 559 kb, 338 kb, and 108 kb, respectively. Most tools had a low PPV of less than 5%, and CoNIFER showed the highest PPV of 10.41%. The number of exons in CNVs was significantly lower in false-negative than in true-positive CNVs (P< 0.001). True-positive and false-negative CNVs had averages of 32.99 and 7.77 exons, respectively. Moreover, the median size of CNVs did not differ between true-positive and false-negative CNVs (156.19 kb and 118.80 kb, respectively).