Gene-based common and rare variant association analysis

Question

[seta]

Dear all,

I’m going work on a given gene with about 170 variants derived from whole genome sequencing of a population (unrelated and related individuals). Most variants are low frequent and rare (AF < 0.05, AF<0.01); in fact there are just 14 common variants and the rest of them are rare. Also, some biochemical variables are available for the participant. I want to find any association between any variants with the available variables (continues variables and also binary variables). As I read the SKAT (SKAT_CommonRare function) is suitable for analyzing both rare and common variants, but it doesn’t consider relatedness and assumed the unrelated subjects. On the other hand, FamSKAT-RC within FamSKAT package is a family-based association test for both rare and common variant, which needs known pedigree. But, I have both related and unrelated samples in the dataset. Besides relatedness, population stratification is another issue here.

Could you share me your suggestion to do an association analysis in this situation (both common and rare variants) considering relatedness and population stratification adjustment?

Thanks in advance

Answer 1

[Kevin Blighe]

Hey,

I would honestly just write the code myself. You could consider a conditional logistic regression model with family as the matching stratum. In R, one can perform conditional logistic regression with clogit. I previously did this in 2016 for a Costa Rican trios study, where I also had to adjust for population stratification (via inclusion of PC1 and PC2) and other covariates. The code for that eventually became a Bioconductor Experiment Package called RegParallel; however, the manuscript is still being published, I believe.

You could consider encoding the variants as categorical (Het, Hom, Ref) or continuous (allele tallies). You could also consider first 'summarising' the rare variants across genes. Obviously, there is no single correct approach - for some genes, a single variant will be sufficient to 'knock out' or disrupt the function of the gene; whereas, in others, it may very well require 'compound' (i.e. multiple) variants, thus increasing the 'burden' of the rare variants on the gene's functionality.

Another idea is to use penalized regression, as these did: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3842118/ (Heather Cordell is an international leader in family-based analyses, and is a pride of the UK genetics community).

Kevin

[seta]

Hi Kevin,

Thank you for your response and introducing your package. Sorry, I read the manual of RegParallel, but it isn't sufficient for me to do the analysis as it's my first experience on this issue. I have the data in plink format, could you please kindly show me how I can test the association of my rare and common variants (within a given gene) with a quantitative variable along with adjusting for kinship (the data belonged to related and unrelated individuals) and adjusting for population stratification using your tool?

Thanks

[Kevin Blighe]

Hey, perhaps you should consider PLINK/Seq: https://atgu.mgh.harvard.edu/plinkseq/assoc.shtml