Figure 6

Sanity check using simulated LC-MS runs. Left: Log-scaled Mahalanobis distances for the simulated LC-MS runs. The black line gives a cutoff for a significance level of 5% as computed from the χ ² distribution. Right: Comparison of Mahalanobis distances computed using standard (y-axis) and robust PCA (x-axis). Good runs are given in green up-pointing triangles and simulated outlier are given in red down-pointing triangles. The vertical and horizontal lines indicate the 5% cutoff for both distances. This plot shows that the robust approaches have advantages if the data contains outlier: using standard PCA we would not classify most of the simulated outlier correctly, furthermore standard PCA classifies several good maps as outlier.