Fig. 4: Evaluation of CS-CORE in recovering co-expressed gene clusters and principal components of co-expression matrices using simulated data, compared to locCSN, Noise Regularization, Normalisr, Pearson correlation of log normalized data (Pearson), propr, Spearman correlation of log normalized data (Spearman), SpQN, ρ-analytic PR and ρ-sctransform.

a Heatmaps of true and estimated co-expression networks from simulations. When plotting results from each method, genes were ordered by applying hierarchical clustering to the estimated co-expression network and color coded by their true gene cluster labels. b Adjusted Rand index (ARI) between true co-expressed gene clusters and clusters extracted from co-expression networks estimated using different methods. The estimated clusters were obtained as described in (a) with the number of clusters set to 4. c Accuracy in recovering principal components, calculated using subspace distance²⁵ between the top four singular vectors of the true co-expression matrix and those of the estimated co-expression matrix. d Spearman correlations between the expression levels and estimated average co-expression levels of genes, with ground truth calculated from simulation settings marked with a dashed line. e ARI between gene clusters extracted from estimated co-expression networks and gene clusters extracted from clustering gene expression levels, with the true ARI calculated from parameters used in simulation settings marked with a dashed line. b–e were evaluated with 25 replications. Data are presented as boxplots (n = 25 per group; center line, median; box limits, upper and lower quartiles; whiskers, up to 1.5 × interquartile range; points, outliers). Source data are provided as a Source Data file.