I can conceive of solid ways to do this, but none of them will be very simple. The general idea is to boil down your vector to a single value (or distribution) and then query all other genes for which you have sufficient data to see how their distribution compares. This is very non-trivial, since you'd want to account for things like gene size, proximity to CpG islands, GC content, etc. that might influence baseline methylation expectation. In practice, I wonder if it'd be simpler to use a random forest trained on all of the other genes to simply predict from them what you'd expect to see in your gene of interest. You could derive a prediction interval from that and compare it to your known value(s).
While i think Devon's RF answer is likely to be the only true/possible way to do this in a sensible way, i'll throw in my ¥0.02 and say that while 'poised' genes do exist, typically methlyation of a gene is almost-0% or almost-100% . If it is not clearly obvious from a sum of your vectors, then id be wary about confidence. A hypomethylated gene but a hypermethlated promoter/enhancer means all bets are off. But really, you need more data - 'Hypo' and 'Hyper' liturally means under/above, so its difficult to make a call without saying what is above/under it :)
If costs are an issue, and you now have a target gene, check out SeqCap Epi: http://www.nimblegen.com/products/seqcap/epi-system/
A LOT cheaper than getting a replicate with WGBS
Sort of depends on the context. Is this "hypo/hypermethylated versus the background average", or in one condition versus another, or...
I have only one healthy individual and I'm wondering whether that gene is hyper of hypo compared to other genes..is there any solid method to do so ?
I can conceive of solid ways to do this, but none of them will be very simple. The general idea is to boil down your vector to a single value (or distribution) and then query all other genes for which you have sufficient data to see how their distribution compares. This is very non-trivial, since you'd want to account for things like gene size, proximity to CpG islands, GC content, etc. that might influence baseline methylation expectation. In practice, I wonder if it'd be simpler to use a random forest trained on all of the other genes to simply predict from them what you'd expect to see in your gene of interest. You could derive a prediction interval from that and compare it to your known value(s).