Figure 3

The role of dinucleotide periodicities in well-positioned nucleosomes. A) Dinucleotide frequencies in well-positioned in vitro nucleosomes. Each curve shows a relative dinucleotide frequency at a given position (with respect to the nucleosome dyad) for the set of well-placed nucleosomes selected from the Rsa I in vitro assay (see Methods). Dinucleotide counts used to calculate the frequencies include both forward and reverse strands for each well-placed nucleosome. We define the relative frequency of a group of dinucleotides as the sum of frequencies of all dinucleotides in that group at a given position, divided by the sum of genome-wide frequencies for the same group of dinucleotides. The groups plotted (with a 3-bp moving average) are AA/AT/TA/TT, CC/CG/GC/GG, and mixed dinucleotides (one A or T and one G or C nucleotide). B) Dinucleotide frequencies in well-positioned in vivo nucleosomes. Same as (A) but using well-placed nucleosomes from the Valouev et al. dataset [39]. C) Predicting well-positioned nucleosomes in vitro. Each curve shows a probability enrichment predicted by a given model at a given distance from well-placed nucleosomes. Probability enrichment is defined as the predicted probability at a given position, divided by the genome-wide mean of the predicted probability profile. Probability enrichment is averaged over all well-placed nucleosomes in the Rsa I in vitro assay; the resulting curves are smoothed with a 7-bp moving average. The two models shown, N = 2 position-independent (magenta) and spatially resolved (green), were fit on the Rsa I in vitro data. D) Predicting well-positioned nucleosomes in vivo. Same as (C) but with models fit on, and well-placed nucleosomes selected from the Valouev et al. dataset [39].