The 23 features used in this study.

1.Values of the two sequence motif features were generated by scanning sequence with Position Weight Matrices from ScerTF [43].
2.p bound or unbound genomic region for a TF in question based on ChIP data
3.The 11 chromatin state features (CS) were obtained from Pokholok et al. [44]. The values for H3, ESA1, and GCN5 are averaged value over the analyzed genomic region.
4.The 10 DNA structure features (DS) were generated from principal component analysis (PCA, see Methods) on 125 DNA structure properties from DiProDB [41]. PC1-5: the average of principle component values over each target genomic region. ΔPC1-5: the difference between the average value of a particular principle component over a target genomic region and the average value of the same principle component over regions flanking the target.
5.The biological meaning was interpreted from top 10 dinucleotide properties having highest PCA loading coefficients.
The 23 features used in this study.