Figure 2

Comparison of Fisher's exact test and Bayesian association score. Scatter plots of Fisher's Exact Test p-value as a function of Bayesian association score. The 2D distributions demonstrate how the relationship between Fisher's Exact Test and our Bayesian score depends on the logOR threshold we use. Each sub-plot represents a different threshold value ranging from 0 to 1 – as indicated by each subtitle. For a particular threshold value, a pixel on the plot represents the local density of miRNA-TF pairs having the corresponding p-value (from the y-axis) and Bayesian Probability (from the x-axis). Here darker shaded regions indicate higher densities. For a 0 threshold, the Bayesian Test and Fisher's Test agree exactly. As we increase the threshold, we see fewer and fewer TF-miRNA pairs that are highly associated as measured by both ranking criteria (pairs whose measures approach 1 at the x-axis and 0 at the y-axis). The higher the threshold, the more emphasis we are placing on the size of the TF-miRNA association (as measured by a log Odds Ratio) and the less emphasis on sample size. Note that a very high Bayesian probability implies that the associated p-value will be small, no matter what threshold we use.