Fig. 2

SVM model performance as a function of top genes using an augmentation strategy of no data augmentation (a) (N = 1260), sampling to the mean class size (b) (N = 1260), or sampling to twice the mean class size (c) (N = 2520). Model performance was averaged over 30 iterations. In strategy (a), the model was trained using the training set of the original data T0 with a 2:1 training-test split stratified by class. In strategies (b) and (c), the model was trained using the augmented training set T2 with a 2:1 training-test split stratified by class. Performance was measured by multiclass AUC in the test set of T0. The p-values in the bottom panel indicate the enrichment of the gene signature in muscle specific genes