My experiment is based on a quantitative continuous variable levels (B12 vitamin) in a sample of patients retrospectively selected based on values that can take (under our criteria):
- low levels: under the physiological or normal reference values but not enough to be considered deficiency
- deficiency (which is associated with poor nutrition, gastric disease, etc.). These data are excluded.
We are interested in low levels. Our hypothesis is to prove if mutations in 10 genes can be associated with these physiological low levels. These physiological low levels are quantified under the format of maximum and minimum value of B12. The genes can take 3 possible categories:
- wt: absence of mutation
- het: half of mutation
- hom: full mutation
- I would like, if possible, to try something like a genetic risk score, if multiple het or hom states in multiple genes can be additive. I don't know how to approach this, because there can be multiple permutations, and maybe permutation 1 has a stronger association with low levels than permutation 2.
- Any ideas on how to carry this out? I don't really know if mutations are additive or opposite. Not sure if some clustering approach can be helpful here, like PCA or euclidean distance, linear discriminant analysis... Any advice is welcome.
My idea is to execute a summary of levels of B12 for each gene and see in a boxplot possible differences, perhaps look for some association (or correlation), but I doubt it.
Any idea or previous input or experiment like this would be tremendously welcome.