Abstract
Background/Objectives: Whole exome sequencing (WES) is an effective method for detecting disease-causing variants. However, copy number variation (CNV) detection using WES data often has limited sensitivity and high false-positive rates. Methods: In this study, we constructed a reference CNV set using chromosomal microarray analysis (CMA) data from 44 of 180 individuals who underwent WES and CMA and evaluated four WES-based CNV callers (CNVkit, CoNIFER, ExomeDepth, and cn.MOPS) against this benchmark. For each tool, we first defined software-specific problematic genomic regions across the full WES cohort and filtered out the CNVs that overlapped these regions. Results: The four algorithms showed low mutual concordance and distinct distributions in the problematic regions. On average, 2210 sequencing target baits (1.23%) were classified as problematic; these baits had lower mappability scores and higher coefficients of variation in RPKM than the remaining probes. After the supplementary filtration step, all tools demonstrated improved performance. Notably, ExomeDepth achieved gains of 14.4% in sensitivity and 7.9% in positive predictive value. Conclusions: We delineated software-specific problematic regions and demonstrated that targeted filtration markedly reduced false positives in WES-based CNV detection.