Table 1 Mean deviation and mean squared deviation. For five datasets and three categories of missing values, imputed values using WeNNI are compared to duplicate controls. The first columns show the number of spots (N), the uncorrected mean squared deviation (msd, defined in eq. 3) and the uncorrected mean deviation (md, defined in eq. 4) relative to random expectation, $\sqrt{\text{msd}/N}$. The sd/rd spots are systematically over/under-estimated, with a deviation much larger than random expectation. For the other missing spots, the deviation varies in sign and is more comparable to random expectation. For sd and rd spots, validation results of a constant correction are also shown, revealing mean deviations well within random expectation, except for the MEC tumours dataset (but there it is noticeably lower than the uncorrected result). The msd is reduced compared to the uncorrected result. The relative reduction is reported in the rightmost column.

From: Accounting for one-channel depletion improves missing value imputation in 2-dye microarray data