MAD Quality Metric

Replicate log-ratio standard deviation

The Median Absolute Deviation (MAD) quality metric is more accurately described as the replicate log-ratio standard deviation. This metric evaluates reproducibility between two biological or technical replicates within RNA-seq/RAMPAGE/Entex experiments in terms of similarity of the expression levels across all pairs of genes with the present cut-off. We measure the standard deviation of the log-ratios between replicates using a robust approach, and the values are calculated per pair of replicates. 

The MAD score between replicates X₁ and X₂ of the sample X is calculated as:

MAD = 1.4826 * median | log₂(X_{1_i}/X_{2_i}) | 

where
X₁ : [X_{1_1},X_{1_2},..,X_{1_N}] expression values in the replicate #1 
X₂ : [X_{2_1},X_{2_2},..,X_{2_N}] expression values in the replicate #2 

The MAD score is calculated on normalized FPKM values for genes G:1..N that have X_{1_i}, X_{2_i} more than preset cut-off >1 FPKM. 

Figure 1. Distribution of the replicate log-ratio standard deviation by the sample type. The dashed red line shows mean values per each sample type; the number of the samples is indicated in the parenthesis of the legend. 

References

A benchmark for RNA-seq quantification pipelines (in press)