Extended Data Fig. 6. The statistical performance of the LASSO–PLSDA model.

To estimate the statistical significance of the model prediction by LASSO–PLSDA, two types of permutation tests were used: (1) shuffling the outcome label and (2) selecting randomized data sets. Each permutation test was performed 1,000 times and the prediction accuracy was calculated based using a receiver operating characteristic curve (shown in the dots). The empirical null distributions for each permutation test were generated and the nominal P values were calculated comparing the true model to the null distributions.