Figure 3.

Reproducibility of four normalization methods. For each of four developmental stages in a C. elegans data set (0, 36, 48 and 60 h, see [10]), the figure shows the median absolute coefficient of variation (MEDACV) for each normalization method, for genes that are primarily Absent or primarily Present in replicate hybridizations. For Absent genes, frequency methods have lower MEDACV than AD methods because of the damping of low-end noise. For Present genes, F^s and AD^s are roughly equivalent and outperform the unscaled methods in all cases. Numbers of replicate hybridizations were three (36 h sample) or four (0, 48, 60 h samples).