the linear relationships observed in Figure 2. Second, we used GC content of 1 Mb window around each marker in the regression model, since this long-range GC content appears to be better correlated with the variation of signal intensities. Third, we used markers that are at least 1 Mb away from each other in the model building to eliminate potential dependence between markers, that is, correlated signal intensities of neighboring markers due to factors unrelated to GC content, such as being covered by the same CNV. Despite these differences, it is clear from all studies that genome-wide GC patterns provide a strong basis for signal adjustment in SNP genotyping platforms.
