Finding evolutionarily conserved cis-regulatory modules with a universal set of motifs – Supplementary Materials

where P (w|M) is the probability of observing w given the motif model (drawn from the frequency matrix) and P (w|B) is the probability of observing w given the background model (estimated from the sequence). All subwords w satisfying LM (w) > tM are classified as M -occurrences. There are two standard approaches to the choice of the threshold tM [1]. The first one aims at restricting the number of false positive motif occurrences. For assumed type I error level α1, tM is chosen to satisfy P (LM (w) > tM |B) = α1. Its disadvantage is poor control on the classification of true M -occurrences. The second approach (setting tM satisfying P (LM (w) < tM |M) = α2 for assumed type II error level α2) restricts the number of false negatives. Unfortunately, it leads to the loss of control on the number of false positives, and consequently to significant disparity in the number of predicted instances of strong and weak motifs (i.e. motifs easily and hardly discriminated from the background). As it is explained in the main text, our method of CRM identification takes into account both positive and negative signals from the promoter sequence. Thus the control of both error types in the motif prediction has to be balanced, in the sense that the number of false positives should be of the order of the number of false negatives. Following the approach proposed by [1] we set the threshold tM satisfying the equation