Fig. 5: Performance comparison of PHet against 16 baseline methods using simulated and HER2 datasets.
From: Heterogeneity-preserving discriminative feature selection for disease-specific subtype discovery
a, b F1 scores of 17 different methods to detect the true DE features in perturbed samples. c, d The number of selected features for each method. a, c The performance of the 17 methods is compared under five scenarios, where the number of perturbed case samples varies from 1 to 9 ({1, 3, 5, 7, 9}). b, d The performance of the 17 methods is compared under five scenarios, which correspond to an increasing number of perturbed samples, from 2 to 18, in both case and control groups. Results (a, b, c, and d) are obtained using ten simulated datasets (Supplementary Table 1). e The average F1 score across 1000 batches of the 17 algorithms in identifying the true 20 biomarkers on HER2 data (Supplementary Table 1). The 95% confidence intervals for F1 scores are provided for each algorithm based on their top k predicted features, with k ranging from 1 to 50. All values are provided as a Source Data file.
