Since Lowess regression has been previously used to reduce genomic waves in array-CGH platforms (4), we investigated its application to SNP genotyping arrays. The method relies on correlating signals from neighboring markers and smoothes the signal intensities continuously along the chromosome. Unlike array-CGH studies, where each CNV may be represented by one or very few clones (probes), the SNP genotyping arrays may reveal CNVs covered by a few SNPs or many hundreds of SNPs. Therefore, the smoothing procedure may not work well for CNVs that are of vastly different sizes. To demonstrate this, we applied Lowess regression with several different window sizes on two samples affected by strong genomic waves (Supplementary Figures 5 and 6). When smaller window sizes are used in the Lowess regression, the wavy patterns in signals are successfully eliminated, but the true CNV also disappears from the adjusted signal intensities. When a larger window size is used, the decreased signal intensity in the CNV region is preserved, however, the wave adjustment no longer works well. Therefore our signal adjustment procedure, which operates on each SNP independent of signals in neighboring SNPs, is better suited for SNP genotyping platforms.
