Fig. 3: Performance comparison of PHet against 16 baseline methods across 6 single-cell transcriptomics and 11 microarray datasets.
From: Heterogeneity-preserving discriminative feature selection for disease-specific subtype discovery
F1 scores of each method for detecting the top 100 DE features that are obtained using LIMMA for both microarray and scRNA-seq (N=17, a), six single-cell transcriptomics (N = 6, d), and 11 microarray (N = 11, g) datasets. The number of selected features by each method using both microarray and scRNA-seq (N = 17, b), six single-cell transcriptomics (N = 6, e), and 11 microarray (N = 11, h) datasets. The adjusted Rand index of each method for both microarray and scRNA-seq (N = 17, c), six single-cell transcriptomics (N=6, f), and 11 microarray (N = 11, i) datasets (Supplementary Tables 2 and 3). The box plots show the medians (centerlines), first and third quartiles (bounds of boxes), and 1.5 × interquartile range (whiskers). A ♢ symbol represents a mean value. A green dashed line indicates the best-performing result on a dataset of PHet on each metric, while a red dashed line represents the worst-performing result on a dataset of PHet on each metric. Dot plots of F1 scores (j), number of selected features (k), and adjusted Rand index scores (l) are presented for each method applied to both microarray and single-cell transcriptomics datasets. All values are provided as a Source Data file.
