# Which statistical test is suitable for my data?

I have biological data which for example looks like following

chromosome.location methylation.level
location_1  100
location_2  0
location_3  0
location_4  0
location_5  100
location_6  50
location_7  100
location_8  0
location_9  100
location_10 0
location_11 100


Methylation.level values are between 0% to 100% in the data file, but biologically methylation level of a chromosomal location in a given time is 0% or 50% or 100%. Both 0% and 100% has big biological meaning. 50% methylation has only little importance.

When I plot the actual data it looks like this:

I have untreated and treated groups.

• Null hypothesis: There is no difference in methylation pattern between two groups.
• Alternate hypothesis: There is difference in methylation pattern between two groups

Which statistical test would be suitable for this sort of distribution?

• Is it actually possible to get true 0s & 100s, or are those rounded, eg? How do these proportions arise? Is it something like the number of heads out of a certain number of coin flips, or is any %age value possible? Do you think the 0s & 100s arise from different processes than the rest? – gung Mar 17 '17 at 12:19
• Yes, it is possible to get 0s and 100s. In two strand of DNA there are 3 cases. 1) Cytosine in one strand is methylated (50% methlation). 2) Cytosine in both strands are methylation (100% methylation). 3) Cytosine in none of the strands is methylation (0%). Methylation is addition of -CH group to Cytosine. Generation of other values apart from 0%,50% and 100% are because of technological limitation. Theses values are not rounded. – World Mar 20 '17 at 6:59
• Using modern sequencing technology, real DNA is chopped into small fragments and those fragments are aligned to human genome. Then percentage of methylated Cytosine in a specific chromosome location is calculated. Please check this link dropbox.com/s/643bglt6cmspvwe/… – World Mar 20 '17 at 7:42
• So it sounds like you have a certain number of fragments, each of which is methylated or not. Is that right? If so, these are binomial data. – gung Mar 20 '17 at 12:43
• yes, each Cytosine in a fragment is either methylated or non-methylated. – World Mar 21 '17 at 8:55