I have studied the methylation pattern of a given gene, and found 3 methylations. My data is given in binary mode - 1 methylated, 0 not methylated. I have an N=8/group and have calculated the average methylation state of each CpG and represent it as a %.
Per each individual, I have calculated the average methylation status of their gene - i.e. if they have 2 CpG methyl out of 3, they are 67% methylated, and then calculate the group average.
I am now interested in correlating my methylation state to other biological parameters (for example, gene expression of this gene in a certain tissue). What kind of correlation test should I carry out and how can I represent it? The way I have my data as mentioned above, I have many 0's and only 4 different numbers: 0, 33, 66 & 100. I only have access to SPSS and GraphPad
Hello Alice
If I understand correctly you want to test the association between your treatments and the level of methylation of your gene of interest.
Generalized Linear Models (GLMs) handle well non-normally distributed data such as counts or proportions and give you several options.
If you know both the successes (methylated sites) and the failures (unmetylated sites) the most appropriate model would be a binomial one. A suitable alternative is a GLM with gamma distribution using proportion of methylated sites as your independent variable. If instead you cannot reliably assess the lack of methylation in your sites of interest you should consider using a GLM with poisson distribution.
All of the above are supported and well documented in R, which is a freely available and open-source statistical programming language. It would surprise me if they were not available for SPSS as well.
Hope this helps
Dear Alfredo,
Thanks for your quick reply, I will give it a try and see how it goes!
Rather than aggregating the CH3 binary scores into 4 categories (0,33,67,100), you could consider each separately and derive 16 categories. Here, you have 4 sites which are either 1 or 0. The recombinatorial categories are 0000, 0001, 0010, 0100, etc. through the 16 permutations possible with 4 character positions and 2 possible chars at each position = 2^4. Not sure how deep the data set is, but the approach would be to test for differences in expression among genes aggregated into these categories.
Also, highly recommend adding R to your 'toolbox'.
If your methylation data is coded as 0 or 1, doing a logistical regression with methylation as the dependent variable and
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS