Skip to main content

Table 1 Operating characteristics for simulated data

From: A Bayesian calibration model for combining different pre-processing methods in Affymetrix chips

  DE Non DE FP (%) TN (%) TP (%) FN (%) sd
Combined 200 800 21 (2.6) 779 (97.4) 179 (89.5) 21 (10.5) 5.0
Method 1 (exp(λ) = 1, ϕ = 1) 200 800 61 (7.6) 739 (92.4) 139 (69.5) 61 (30.5) 5.5
Method 2 (exp(λ) = 0.5, ϕ = 0.5) 200 800 80 (10.0) 720 (90.0) 120 (60.0) 80 (40.0) 4.8
Method 3 (exp(λ) = 0.5, ϕ = 2) 200 800 26 (3.2) 774 (96.8) 174 (87.0) 26 (13.0) 4.1
Method 4 (exp(λ) = 2, ϕ = 0.5) 200 800 122 (15.2) 678 (84.8) 78 (49.0) 122 (61.0) 3.3
Method 5 (exp(λ) = 2, ϕ = 2) 200 800 45 (5.6) 755 (94.4) 155 (77.5) 45 (22.5) 5.8
  1. The table presents the number of False Positives (FP), True Negatives (TN), True Positives (TP) and False Negatives (FN) in the first 200 genes ranked accordingly to their tail posterior probability. Note that FP = FN since the size of the list of differentially expressed genes is equal to the number of true positives. The combined method shows the smallest number of FP and FN. Out of the 5 pre-processing methods, the one characterised by a relative bias parameter ϕ j > 1 and a variability parameter exp(λ j ) < 1 (Method 3) shows the best performance, due to the combination of high signal and low variability around the mean gene expression. The standard deviation calculated on the 10 runs is also reported in the table. Note that as FP, FN, TP and TN are a linear function of each other given a fixed sum, the standard deviation is the same for the 4 variables.