HMM training for PMD command

[Juheon Maeng]

Hi again,

I ran pmd for two human samples with distinct global methylation level: sample A: higher beta,  sample B: lower beta. The sample B is the same cell line of sample A but is treated with DNMT inhibitor. Considering drug-induced hypomethylation in sample B, I expect to see the total size of PMDs is bigger in sample B but the result was the opposite. 

I looked into a couple of PMDs that are unique to sample A. Even though sample A-specific PMDs showed higher methylation levels than corresponding regions in sample B, those regions are called as PMDs in sample A but not in B. 

I think it could be due to sample-specific HMM parameter training (ex. hypomethylated samples, HMM will call bins as PMDs at higher methylation level compared to hypermethylated samples) Could you explain a little more about how does the PMD train HMM parameters? 

Best regards,
Ju Heon Maeng