Ten-fold cross-validation was performed on the selected 2,504 non-trio samples.
Samples were divided into 10 batches and stratified by superpopulation (EAS, EUR, SAS, AFR, AMR) to ensure balanced representation. In this study, superpopulations are treated as populations.
4 batches of 251 samples
6 batches of 250 samples
In each fold:
90% of data serves as the reference panel.
10% of data serves as the target set for imputation (using to prepare true VCFs and downsampled/psudo-array inputs).