We will eventually find that there's the equivalent of JPEG encoding (plus SSL and ECC) and that today we're doing the equivalent of someone with no knowledge of SSL or JPEG staring at a tcpdump trace of a browser loading an image over https, trying to make sense of the resulting hex dump.
Trying to figure out how a single compressed/encrypted buffer works given only it and its resulting image is certainly an unwinnable task, but increase the sample size to a few million or billion and I don't see why the algorithm couldn't eventually be cracked.
If genes are the data, then methylation is more like an additional piece of metadata directing the cell (potentially different across cells) how to use the data. One possible use for this metadata is potentially error correction, but it does much more than that.
