SNP - Protein interaction

[mvh]

Hello there!

I am a PhD student with a rather complex problem.
Since my specialty is molecular biology and genetics, I am having a rather hard time with a statistical issue, and is hoping that the experts within this forum can help me out

This is what I would like to do.
I have several protein measurements from differnt individuals in a case-control study, and a SNP genotype panel. I would like to see if there is any association between SNP genotype and protein levels, between cases and controls. This would be quite simple if my protein measurements was normal distributed, but unfortunately they arent, and they cant be manipulated to be normalized. Does anyone in here have an idea on how to proceed. I would rather not "group" the protein levels...

Hope you can help me out.

Thanks!
Mads

[TheEcologist]

Now I'm not quite sure how to help you (or if I can), I'll need a data description first.

How are your SNiP's quantified?
How are your protein levels quantified?

Categorical, ordinal, continuous ect ect

Secondly how is your response distributed then?

This would be quite simple if my protein measurements was normal distributed, but unfortunately they arent

What would you do if your protein measurements were normally distributed?

It of course depends on all of the above, but if you were thinking of running an regression or ANOVA type analysis remember that only your residuals must be normally distributed. The error around the model should be normal, the response itself not (necessarily).

[mvh]

How are your SNiP's quantified?
The SNP are genotyped and presented as AA or AB (heterozygote) or BB, so catagorically

How are your protein levels quantified?
The protein levels is continous, and unfortunately the residuals is not normally distributed, they seems more "randomly distributed" (Please forgive my lack of statistical slang or technical words)

My first thought was to use ANOVA to do the analysis, but since the lack of normally distribution I couldnt.

I could chose to set some criterias for the protein concentration measurements, and analyse it in concentration groupings, but I would rather not.

I really hope you can help me out, and appreciate your fast response!

[TheEcologist]

The non-parametric alternative to an ANOVA is the Kruskal-Wallis rank sum test.
You can try it, it is less powerfull though.

Also is it possible to post boxplots of your data per case or do you have multiple cases?

[mvh]

The non-parametric alternative to an ANOVA is the Kruskal-Wallis rank sum test.
You can try it, it is less powerfull though.
I already done KW testing to see if there was a direct correlation between genotype and protein concentration, unfortunately I cant implement case/control status into this analysis, and thats what I am looking for

Also is it possible to post boxplots of your data per case or do you have multiple cases
I am not sure what you mean, I have several controls and cases

[TheEcologist]

Oke so it indeed seems that you have gotten yourself a complex problem

You know I feel like I’m still guessing here as I still don’t know the details of your study (how many cases ect). This type of Complex problems will require more info.

But in light of what I do know and what I suspect you want to do: I would start out with a modelling study.

(Maybe you have already done this but) Look at boxplots per case/control situation see if there is a difference or a detectable trend (something that might tell you how the relationship is ergo what potential model candidates are). Secondly try to figure out how the error if this model would look like ( we know its not normal but could it be then e.g. poisson or gamma distributed or ….).
Once you have a better idea about this then look into GLM’s (generalized linear models), secondly maybe GAMS (http://www.statsoft.com/textbook/stgam.html). I recommend borrowing Statistics: An Introduction using R by Michael J. Crawley (it’s the most comprehensible description of GLM’s I’ve seen so far).