Average imputation accuracies of the eight methods, fastPHASE (fPH), NN, WNN, SVM, Npute, MC, NeuralNet (NeuN) and BaseLine (BL), on the three different density level mouse datasets: density-0.01, density-0.1 and density-1. Compared to the above missing genotype imputation, the general tendencies are different: 1) in all three density level datasets, our methods NN and WNN performed no differently from fastPHASE, and even slightly better on the density-0.01 datasets; all three of them performed statistically significantly better (by ∼6%, p-values <0.0001) than previously the best method Npute. 2) There are only three groups here, the first group consists of fastPHASE, NN and WNN; the second group includes SVM, Npute, MC and NeuralNet; and the last group contains only BaseLine. 3) Even on the low density datasets, all other methods performed better than BaseLine; on the median and high density datasets, the gaps became significantly larger. It should be pointed out that SVM performed very strangely on the high density datasets with missing rates 5% (and 10%, 20%). We have in fact separately tested multiple times, but the same pattern was always there.