We investigated the effects of using different CNV calling algorithms and found that the choice of analysis tool can be as important as the choice of array for accurate CNV detection. Different algorithms give substantially different quantity and quality of CNV calls, even when identical raw data are used as the input. This has important implications both for CNV-based, genome-wide association studies and for the genetic diagnostics field. We show that algorithms developed specifically for a certain data type (e.g., Birdsuite for Affymetrix 6.0 and DNA Analytics for Agilent data) generally perform better than platform-independent algorithms (e.g., Nexus Copy Number) or tools that have been readapted for newer versions of an array (e.g., dCHIP on Affymetrix 6.0 data).
