Figure 4

Accuracy versus drop rate for various combinations of preprocessing and regression analysis for predicting k . Results are shown for the Omni1-Quad (A), Omni2.5-Quad (B) and Omni5-Quad (C) platforms. The combination with the highest accuracy is quantile normalization coupled with regular logistic regression (solid gold line). Other combinations, such as quantile normalization with loess adjustment and ordered logistic regression (dotted blue line), or quantile with ordered logistic regression (dashed red line) and quantile with loess adjustment and regular logistic regression (dashed green line) are either generally less accurate, or perform inconsistently between platforms. Accuracy (y-axis) refers to concordance with independent HapMap calls and drop rate (x-axis) is the proportion of calls removed due to low call confidence as measured by silhouette width.