Fig. 2

Selecting the best HMM and defining genomic regions for each of HMM states. A We trained 50 HMM models (other 25 models were shown in Additional file 1: Table S4) with different numbers of initial states and select the best model with the smallest BIC score (the highlighted model). B A line plot showed seven states are redundant in the current “best” model. We applied the second round HMM training by removing those seven redundant states. C The transition probabilities of the final 13-states HMM. The transitions were from states on y-axis to the x-axis. D The mark-state probabilities that derived from the emission probabilities. Each column represents a histone mark and each row represents a HMM state. E A distribution of each of HMM states in 100Kb Upstream TSS. F A distribution of each of HMM states in the gene body