Figure 1

Schematic illustration of the differences between the ratiometric analysis and correlation metrics. A) For five samples (A-E), the expression levels of genes 1–4 are measured (top graph). The box on the right shows the analysis of expression relationships using a Pearson correlation. Only gene pair 1:2 is identified as a significant interaction, (R² = 0.99). In contrast, the ratiometric method (box on the left) identifies both pairs 1:2 and 3:4 as significant. The PE method does not capture the second relationship (3:4) as the FPKM ranges of the two genes are too narrow for a regression line to be stable. On the other hand, the RA model assesses only the FPKM fold-change across samples, is much less sensitive to narrow FPKM ranges, and identifies both pairs. B) Shown is simulated expression data for two genes, A and B. The expression levels of A were generated from those of B as follows: a _i  = 2b _i  + u _i . For each dataset, the expression range of B was varied by increasing CV(B) from 0 to 25% of the mean level of B(μ _B , = 500). The expression level of B is thus normally distributed B ∼ N(500, % B). For each value of CV(B), 10 datasets with 100 samples each were generated. The Pearson and Spearman R², the entropy, mutual information, and the CV(A/B), CV(B/A) and Δ _CV -values were calculated for each dataset, and the mean and standard error are shown. Note that the gene pair association does not change along the x axis and the expression of gene B can be used to predict the expression of gene A equally well in all runs. As the expression range for B narrows, the Pearson and Spearman R²-values decrease, along with the mutual information index. In contrast, the ratiometric CV is constant and the relationship between the expression levels of the two genes is always recovered.