slope = 0.52
slope = 0.64
slope = 0.76
slope = 0.74
slope = 0.47
slope = 0.49
slope = 0.63
slope = 0.57
slope = 0.59
slope = 0.49
slope = 0.8
slope = 0.49
slope = 0.6
slope = 0.36
slope = 0.71
slope = 0.67
slope = 0.59
slope = 0.53
slope = 0.44
slope = 1.09
slope = 0.91
slope = 0.6
slope = 0.72
slope = 0.74
slope = 0.52
slope = 0.62
slope = 0.62
slope = 0.74
slope = 0.46
slope = 0.79
slope = 0.52
slope = 0.91
slope = 0.54
slope = 0.71
slope = 0.84
slope = 0.71
slope = 0.69
slope = 0.77
slope = 0.55
slope = 0.59
slope = 0.72
slope = 0.49
Figure S1. Regression of 42 target genes on to the mean Ct of reference genes. Only genes without
undetectable values from the RA SAB dataset was used. X-axis is the mean Ct values of the reference
genes. Y-axis is the Ct values of a target gene.
0.4
-0.2
0.0
0.2
FC.rg
0.6
0.8
1.0
1.2
Fold Change Estimates
-0.5
0.0
0.5
1.0
1.5
FC.dCt
Figure S2. Comparison of fold change estimates from dCt and per-gene regression normalizations in a
simulation study. The simulation was conducted as described in the Supp. Table 3 legends. The
parameter setting are: sample size (n)=20, regression coefficient (b) = 0.58, and variation (sd) =0.2. The
fold change estimates from the regression normalization (FC.rg) are well separated in the three
simulated group means, 0, log2(1.5) and 1og2(2). The estimates from dCt normalization (FC.dCt) have
much more variability around the simulated values than those from the per-gene regression
normalization (FC.rg).