Figure 5:

Our two-phase validation strategy. The first cross-validation happened in the discovery cohort, which was split in five folds of equal sample sizes (Test i) for five cross-validations. In each cross validation, the set ”Test i” (dark orange box) was used for evaluation and the rest four folds (light orange box) were used for training ”Model i” (blue boxes). In the second phase of validation, each trained ”Model i” (blue box) was applied on the completely-unseen replication cohort (green box) to evaluate accuracy and generality.