PMD is a methylation feature that is not well defined. There are many ways of calling it. The most convenient one should be MethylSeekR.

If you are interested,

1. In the paper where the concept of PMD was first introduced:

   A sliding window approach was used to find regions of the genome in IMR90 that were partially methylated, based on the measurements of the level of methylation at each mCG. A window size of 10 kb was used, progressing 10 kb per iteration. When a 10 kb window was identified that contained at least 10 mCG, each covered by at least 5 MethylC-Seq reads, for which the average methylation level of these mCG was less than 70%, the region was extended in 10 kb increments. Extension was terminated when a 10 kb increment was reached that had an average methylation level of greater than 70% or less than 10 mCG, and the region was reported as a PMD.

2. Also, HMM-based approach described in this paper.

3. In one most recent paper, random forest based approach

It's the confusion time. In the field of DNA methylation, we created so many conceptions, however, we actually do not have the precise definition, functions and the code.

Well, partially methylated just means somewhere between the minimum and maximum (though we'd normally set some threshold like "in the middle 50%"). One question becomes how the methylation scores relate to absolute methylation. For example, if the scores end up being similar to the logit of the absolute score, then you'd filter at different threshold than if they were just an approximate linear mapping. Your task, then, will be to try to determine this, likely with a published dataset that also has paired bisulfite sequencing samples, and then filter accordingly.