calling subcompartment for one sample with no replicates

[sotuamax]

Hi,

I am wondering if I can call subcompartments for one sample only (no assignment of --diffdir) after running --pcatype "cis" and "select". I see in the paper, it used CALDER to run for subcompartments. Is there a way I can do it using dchic only?

Thank you so much for your help.

Best,

[ay-lab]

Hi,

dcHiC expects at least two samples. If you're interested in the sub compartments from only one sample, you can use the same sample as a dummy as a second sample in a different name to do the job.

[sotuamax]

Hello,

Thank you for your response. I tried by giving more than one sample in the input file and run "--pcatype subcomp" and it worked. As I am interested in the association between sub-compartments and histone modification markers, for example, I observed the H3K9me3 signal is getting stronger along with stronger B compartments (from B0 -> B1 -> B2 -> B3). The default setting for the number of subcompartment is 6, however, I am not sure if this default number makes sense for my data and always applicable for all the samples I have. So, my question is: is there a way for me to decide a reasonable number for sub-compartments?

Thank you so much.

Best,
Meiyuan

[ay-lab]

If I am not wrong, the "6 state" subcompartment was first shown in Rao et al. 2014 Cell. Thereafter, everyone else started to stick to this number (or close to that number). We also decided to go ahead with this number of states. Finding the optimum number of subcompartments is difficult and deserves a separate line of investigation by its own merit. I don't have a good answer to your question. However, if you have independent signal(s) like K9me3 and/or others you can perform a simple K-means clustering on a M X N matrix (M = Hi-C bins, and N = features like normalized K9me3 signal, compartment scores etc) and perform an elbow method to determine best number of clusters, it can give some estimate about the number of subcompartments that you may need. You can choose the number of subcompartment by setting the --subnum option in dcHiC.

[sotuamax]

Hello,

Thanks for your suggestions. I do have another question related to subcompartment. I noticed that for some of the strong "A" subcompartment (e.g., A2), in the pc analysis, the score is very negative (--pcatype select). Could you explain why we see the discrepancy between A/B analysis and subcompartment?

Best,

[ay-lab]

Is it surrounded by strong A-compartments? Sub-compartment calculation uses HMM, so the surroundings can have affect.