A random forest-based framework for genotyping and accuracy assessment of copy number variations

作者：Xuehan Zhuang, Rui Ye, Man‐Ting So, Wai-Yee Lam, Anwarul Karim, Michelle Yu, Ngoc Diem Ngo, Stacey S. Cherny, Paul Kwong‐Hang Tam, María-Mercé García-Barceló, Clara Sze-Man Tang, Pak C. Sham · 发表于：NAR Genomics and Bioinformatics · 年份：2020 · DOI：10.1093/nargab/lqaa071 · 被引用次数：14 · 研究领域：Genomic variations and chromosomal abnormalities、Cancer Genomics and Diagnostics、Genomics and Rare Diseases

Abstract Detection of copy number variations (CNVs) is essential for uncovering genetic factors underlying human diseases. However, CNV detection by current methods is prone to error, and precisely identifying CNVs from paired-end whole genome sequencing (WGS) data is still challenging. Here, we present a framework, CNV-JACG, for Judging the Accuracy of CNVs and Genotyping using paired-end WGS data. CNV-JACG is based on a random forest model trained on 21 distinctive features characterizing the CNV region and its breakpoints. Using the data from the 1000 Genomes Project, Genome in a Bottle Consortium, the Human Genome Structural Variation Consortium and in-house technical replicates, we show that CNV-JACG has superior sensitivity over the latest genotyping method, SV2, particularly for the small CNVs (≤1 kb). We also demonstrate that CNV-JACG outperforms SV2 in terms of Mendelian inconsistency in trios and concordance between technical replicates. Our study suggests that CNV-JACG would be a useful tool in assessing the accuracy of CNVs to meet the ever-growing needs for uncovering the missing heritability linked to CNVs.