Model scores for the cluster in Figure 2a. Each graph shows how well one of the submodels of PARma matches when aligned to the seven-mer that starts at the corresponding position. For instance in (a), the maximal value belongs to the seven-mer TGCTGCT and indicates that all observed and unobserved T to C conversions match very well, when TGCTGCT is the miRNA seed site. A miRNA targeting the seed site CACATTG (corresponding to the secondary peak upstream of TGCTGCT) is also likely to lead to the observed conversion. The cleavage scores in (b) and (c) indicate how likely the observed RNase Tl cleavages are, given the seed site is at the corresponding position. Both submodels would allow seed sites to start within a small window of about 10 bases and indicate that the secondary peak from (a) is unlikely to correspond to the true miRNA seed site. However, they agree with the primary peak of the conversion scores. Finally, the k-mer activity scores in (d) indicate how many other PAR-CLIP clusters are likely to be explained by the corresponding k-mer and they also point to the seven-mer TGCTGCT. This is indeed the seven-mer-m8 seed site for miR-l5a, and it has been experimentally validated that it targets this cluster .