I have implemented an analysis with ChromHMM, with many histone marks as input. I used a range of different numbers for states and assessed the output of each.
for i in {1..10}
do
java -mx1600M -jar ~/ChromHMM/ChromHMM.jar LearnModel -p 6 ChromHMM_binary ${i}_e $i mm9
done
Across all of the number of states that I assigned, I cannot seem to find a way to compare and find out which of these is the best number that suits the given data the more than all others...... I am new to HMMs and having read the paper: https://www.nature.com/articles/nprot.2017.124 I cannot find a way to assess the strength of the model in ChromHMM. Looking at the 'emissions_no.png' which tells which histone marks are prevalent in each state, I see that none of the histone marks are very strong in any of the given states (the colours that indicate a given histone marks' strength in state 'x' are showing little or no intensity)... this may mean that I have bad data (which is very possible as I am only playing around with the software) but I would like to be more sure and verifiably say that the model is bad or good...