Copy Number Variation Accuracy in Genome-Wide Association Studies

Abstract

Background/Aim: Copy number variations (CNVs) are a major source of alterations among individuals and are a potential risk factor in many diseases. Numerous diseases have been linked to deletions and duplications of these chromosomal segments. Data from genome-wide association studies and other microarrays may be used to identify CNVs by several different computer programs, but the reliability of the results has been questioned. Methods: To help researchers reduce the number of false-positive CNVs that need to be followed up with laboratory testing, we evaluated the relative performance of CNVPartition, PennCNV and QuantiSNP, and developed a statistical method for estimating sensitivity and positive predictive values of CNV calls and tested it on 96 duplicate samples in our dataset. Results: We found that the positive predictive rate increases with the number of probes in the CNV and the size of the CNV, with the highest positive predicted rates in CNVs of at least 500 kb and at least 100 probes. Our analysis also indicates that identifying CNVs reported by multiple programs can greatly improve the reproducibility rate and the positive predicted rate. Conclusion: Our methods can be used by investigators to identify CNVs in genome-wide data with greater reliability.

A Genome-Wide Association Study to Identify Potential Germline Copy Number Variants for Sporadic Breast Cancer Susceptibility

“…[Thompson et al, 2011;Bush and Moore, 2012]; the higher odds ratios, if confirmed in independent cohorts, may support the general observations that CNVs associated with phenotypes confer a higher effect size than the SNP-based associations [Lin et al, 2011]. Common CNVs identified in this study (online suppl.…”

“…However, we refrained from such assumptions, and as such many of the identified CNVs in our study or their potential concordances may have been underestimated in our analysis. This is true of any algorithm comparisons implemented thus far for CNV association analysis, as acknowledged by others [Grayson and Aune, 2011;Lin et al, 2011;Zhang et al, 2011;Cantsilieris et al, 2014;Castellani et al, 2014].…”

“…We therefore made a systematic effort to compare signatures based on criteria of concordances of CNV calls between algorithms (described above). Previous investigators used an arbitrary cut-off of 25-50% of overlap of the sequence as a criterion for concordance [Lin et al, 2011;Zhang et al, 2011;Castellani et al, 2014]. However, we defined concordance as 100% overlap of CNV regions to be on the more stringent side to accommodate the mismatch of default parameters for segmentation by individual algorithms.…”

“…The catalogues of variants available as a source of reference are of limited utility, and a gold standard reference is still evasive, contrary to the previous attempts to create such a reference database Korn et al, 2008;Conrad et al, 2010b;Zhang et al, 2011]. High-throughput methods are not available to validate these signatures in a cost-effective manner, and qRT-PCR is one suggested approach to select a number of targets for a limited sample size [Lin et al, 2011]. Keeping these limitations in mind, several recent studies made recommendations to use more than 1 CNV calling algorithm to ascertain CNV calls [Lin et al, 2011;Zhang et al, 2011].…”

“…High-throughput methods are not available to validate these signatures in a cost-effective manner, and qRT-PCR is one suggested approach to select a number of targets for a limited sample size [Lin et al, 2011]. Keeping these limitations in mind, several recent studies made recommendations to use more than 1 CNV calling algorithm to ascertain CNV calls [Lin et al, 2011;Zhang et al, 2011]. We therefore designed our study to include 2 CNV calling algorithms, Partek Genomics Suite 6.6 (Partek GS) and Nexus Copy Number 6.0 (Nexus CN), respectively.…”