Purchase this article with an account.
Y.-J. Li, A. Dellinger, M. Seielstad, L. Goh, T. L. Young, S. M. Saw; Evaluation of a Novel Trimmed Mean Based CNV Detection Method Using Whole Genome SNP Arrays From Myopia Samples. Invest. Ophthalmol. Vis. Sci. 2008;49(13):2597. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Methods for detecting copy number variants (CNVs) using genome wide single nucleotide polymorphism (SNP) arrays are relatively new and their performance is uncertain. We compared the performance of six CNV detection methods including our proposed trimmed-mean based method (TrimMean) using 26 buccal, saliva, and blood DNA samples from the Singapore Cohort study Of the Risk factors for Myopia (SCORM). We also examined the impact of different DNA sources on the discovery of CNVs.
Twenty-six samples from 13 SCORM participants were genotyped using Illumina 550K SNP arrays. Thirteen DNA samples were from buccal swabs, 7 from saliva, 1 from blood, and 5 from amplified buccal DNA. Six CNV detection methods were evaluated: circular binary segmentation (CBS), gain and loss of DNA, CNVFinder, dChip, cnvPartition, and TrimMean. TrimMean creates a distribution of non-CNV SNP Log R ratios from all samples, where non-CNV SNPs were obtained from the inter-quarter region in every sample. SNPs with intensity outside of upper or lower 2.5% threshold of this distribution are declared as CNV SNPs. A CNV is defined for a segment with at least three consecutive CNV SNPs for all methods. All CNVs were compared to the Database of Genomic Variants and HapMap. Sensitivity, specificity, and kappa values were compared among methods. TrimMean CNV frequencies were compared between myopic (8) and normal (5) samples.
Of the DNA sources, amplified buccal DNA has the highest number of CNVs across 6 methods. For all 13 buccal DNA, CNVfinder is the most conservative method with the fewest CNV SNPs detected (average 327) and CBS detected too many (average 51976). TrimMean detects an intermediate number of CNV SNPs (average 37 338) among all methods. Sensitivity is consistently low (0 to 0.53) due to the liberal assumption that all CNVs in the database are present. TrimMean has an average kappa value of 0.85, slightly below CNVfinder (0.89) and CNVpartition (0.88), but with higher sensitivity. TrimMean found
Amplified buccal DNA did not perform well. Overall, TrimMean is an efficient and robust method to detect CNVs, and parameter adjustments may improve performance.
This PDF is available to Subscribers Only