Epi-Genetics and DNA Computing

I am busy making plans for a long-distance move and so I will have only limited time to put up a stub where questions go today. Will attempt to flesh this out later.

At the local Eugene Science Pub event, a Professor from Pacific University spoke about the concept of nature and nurture in the context of recent work in epigenetics. The topic itself is fascinating and in a nutshell studies the effects of how methyl-groups (CH3) are tagged onto the DNA sequence to disable gene expression (and presumably protein synthesis) as well as tagged onto the proteins around which the DNA sequences are coiled around. In the latter case, the methyl-groups causes the DNA strands to be less tightly wound, exposing them for expression (the reverse of the former case).

On this topic, as a computer scientist and an engineer, I have several vectors of interest:

1) The methyl-groups according to her, are tagged through the aid of enzymes. If I remember my microbiology correctly, this should be similar to the way proteins are constructed - that one part of the enzyme should bind with the methyl-group while the other latches on to other identifier markers on the DNA strand or the proteins. The question really is "how?". I would like to understand the mechanics of it. Surely, this process has to be statistical, right? How would a consistent template of gene expression toggling be achieved by the mechanics of methyl-group tagging across the entire human genome and all the cells in our body? She was talking about how the tags change the regular nature of dominant and recessive gene interplay. Also, even if one assumes some statistical modeling, how do the enzymes tag, so cleanly, some set of specific genes? Do those enzymes identify an entire sub-sequence for tagging? How does the methyl-groups chemically prevent the usual protein synthesis process from working?

2) Now for the stuff I am really excited about (in the long term). Can I make use of this information to design and construct more complex scenarios using methyl-groups to enable yet another model (maybe one that is more flexible) for DNA-based computing? Can I model an environment where I do not have to mimic biology? All I need to do is to respect the chemistry and statistics involved, correct?