Figure 5

Robust performance of RSPR-NR. Simulated data sets with different levels of signal complexity (A) and noise variance ratio (C) were treated with RSPR-NR and PCA, and the distributions of the resulting noise RMS ratio in 40 simulated data sets are plotted. For PCA, the number of top PCs to keep was determined for the minimal noise RMS ratio in each simulation, by the procedure illustrated in Figure 3B. The distributions of the medians of the numbers of PCs kept in RSPR-NR and the distributions of the optimum number of top PCs kept in PCA for (A) and (C) are shown in (B) and (D), respectively. Four FDR conditions of 0.01, 0.0316, 0.1, and 0.316 were used in RSPR-NR and indicated as α 0.01, α 0.03, α 0.1, and α 0.3. The signal pattern conditions used in (A) and (B), which are indicated at the top of the plots, are 10 and 20 (pattern 1), 20 and 40 (pattern 2), 30 and 60 (pattern 3), and 40 and 120 (pattern 4) large and small blocks. A noise variance ratio of 0.01 was used in (A) and (B). The noise variance ratios used in (C) and (D), which are indicated at the top of the plots, are from left 0.00316, 0.01, 0.0316, and 0.1. For the signals, 30 and 60 large and small blocks (pattern 3) were used in (C) and (D). For all panels, a subset row number of 200 and a repeat number of 20 were used in RSPR-NR.