Figure 2

Quality control metrics using the coefficient of variation (CV). (A) Boxplot of the CVs between the raw intensity values for the four probes mapping to a given probe set (CVwithin), computed for all 595 common mouse miRNA probe sets and for all 16 arrays of the lung dataset, excluding the “absent” detection calls (see Additional file 2 “Supplementary Results”). (B) Boxplot of the CVs between the probe set normalized intensity values from the four biological replicates of a given treatment group (CVbetween), computed for all 595 common mouse miRNA probe sets and for all four treatment groups of the lung dataset, excluding the “absent” detection calls. (C) Boxplot of the ratio between the probe set residual variance and the corresponding modeled treatment response (CVtreat), computed for all 595 common mouse miRNA probe sets for the lung dataset, excluding the “absent” detection calls. (D) Boxplot of the CVs between the probe set normalized intensity values from all 16 arrays of the lung dataset, computed for the 10 Exiqon spike-in control probe sets. The boxplots for each preprocessing pipeline (described in Additional file 1: Table S1) show the values relative to the range of the corresponding CVbetween distribution, which is given by the whiskers in panel B and the interval [0,1] in panel D (relativeCVspike, see the “Coefficients of variation” section in “Methods”). Median, red line; first and third quartiles Q1 and Q3, blue box; distribution range given by the two extreme values within the interval [Q1 − 1.5×(Q3 − Q1), Q3 + 1.5×(Q3 − Q1)], black whiskers; outliers, red crosses outside the whiskers.