Figure 4

The framework of Chromia. (1) Data preparation. Chromia takes binned signals of PSSM scores and eight histone marks in the entire genome as input. (2) Training data. Regions centered at TSS and p300 binding sites were selected to train HMMs for promoters and enhancers, respectively. The entire chromosome 1 was used to train the background model. (3) Model training. Three HMMs with a left-right structure and a mixture of Gaussians were trained for promoters, enhancers and background, respectively. (4) Whole genome scanning. Two log-odd scores were calculated for each bin in the entire genome using the trained HMMs. (5) TFBS predictions. Log-odd scores of adjacent bins were averaged to smooth the curve. Bins with a log-odd score greater than other binding sites within ± 2,000 bp were predicted to contain the TFBSs. See Materials and methods for details.