Probe set residual variance (defined as alpha) is independent of signal intensity. (A) A linear regression was performed on α
as a function of signal mean. A slope of approximately zero, and no obvious pattern of residuals, indicates that α
has corrected for all intensity related variance. Note that more data points are available at lower intensity signals, as shown in Fig. 4. (B) Alpha's variance is independent of signal intensity. We defined 'alpha outliers' as having an α
in the top 10 %, and calculated the frequency of 'alpha outliers' in different expression bins, formed by ranking all genes for their signal intensity. The outlier rate remained roughly constant at the expected frequency of 10 % regardless of expression level. The analysis was performed based on the 12 650 probe sets represented in chip U95A, in three paradigms.