Fig. 2. Accuracy of the genotype-correction methods at various MCC fractions.

Machine-learning methods were trained with a “leave-one out” strategy. Each curve consists of twelve data points corresponding to accuracy of the method applied to “contaminated specimens” at various MCCs (the contamination values used are as given in section “Contaminated trio generation”). The accuracy is calculated on the intersection of the calls made on the virtual fetal specimen and the calls made on the real-world child.