P_CT, the probability of conserved targeting as described in Friedman et al., 2009, has been calculated for all highly conserved miRNA families.

To control for site type and sequence features, we calculated a signal-to-background ratio (S/B) for each site at each site's branch-length. For the purpose of evaluating individual sites, assessing controls at each branch length instead of at each branch-length cutoff is necessary in order to avoid crediting poorly conserved sites for having the same sequence as many highly conserved sites. We then converted this S/B to a probability of preferentially conserved targeting (P_CT), which is approximately equal to (S/B - 1)/(S/B) (or near zero, for sites with S/B < 1). This score reflected the Bayesian estimate of the probability that a site is conserved due to selective maintenance of miRNA targeting rather than by chance or any other reason not pertinent to miRNA targeting, allowing for uncertainty in the S/B ratio. P_CT reported on this web site were also smoothed as described in the Supplemental Discussion of Friedman et al., 2009.

Predicted targets of a miRNA family can be sorted by decreasing aggregate P_CT, as described in Friedman et al., 2009. Since P_CT refers to a probability, the aggregate P_CT is calculated as

1 - (   (1 - P_CT)_site1   x   (1 - P_CT)_site2   x   (1 - P_CT)_site3 ...   )

Since overlapping sites cannot be occupied at the same time, some are removed from the table of predicted targets to create a set of non-overlapping sites while maximizing the total context score. These non-overlapping sites are used to calculate the aggregate P_CT. All sites are still shown on the gene-centric detail pages.