Figure 2

WeNNI is most accurate over all ranges of spot quality. The contribution to MSD for specific SNR for the different imputation methods applied to the three different data sets using β = 0.3. These plots were created using a sliding window containing 1% of all spots. Spots with small SNR (low quality) have the largest impact on MSD. (A) In the breast cancer data a weighted scheme is clearly essential and WeNNI is most accurate over all ranges of SNR (B) In the melanoma data a weighted scheme is clearly essential and the weighted reporter average show best performance for an SNR range 0.2–1. (C) In the mycorrhiza data the breakdown of the average reporter methods is very prominent. For the SNR range 0.07–0.4 it is even better to use no impute (green line) than the average methods. The breakdown of the reporter average methods are discussed in the text.