Hi,
I have sequenced exomes of case and control and generated the SNP file. For the gene of my interest I can see SNPs with both cases and control. Now, to confirm whether, this gene is responsible for the disease, what sort of statistic I need to use considering I have multiple rare variations both in case data-set as well as control data-set for that particular gene.
A quick note: There does not currently, nor will there likely ever (unless we're counting StarTrek-like holodeck simulations as a statistical test...in which case I guess perhaps someday), exist a statistical test that will determine whether a SNP is causative. You can use statistics to demonstrate significant enrichment/association but not causation, which will require some bench work and possibly making a mouse model.
Anyway, the most common test to see if a SNP is significantly associated would be a Fisher's exact test. The R function is fisher.test().
•
written
by
Devon Ryan
Thanks Devon for the explanation and the suggestion. I also want to ask you one additional query. If a gene has 7 SNP both in case dataset and control dataset; but the number of individual is different; then how should I do the association study. I think I cant do it SNP by SNP.
Am sorry for asking such basic question; but I am very new to human genetics.
By "the number of individual is different", do you mean between cases and controls or by SNP even within the cases and controls? Having different numbers of cases and controls is extremely common and not an issue at all (usually you have a LOT more controls than cases, since controls are easy to come by). Depending on the exact nature of the disease and data, you can sometimes test things as a group, as we did here (see supplemental table S4). You could also test SNPs individually, but that often makes more sense when you're looking at complex diseases and I assume you're working on a more Mendelian disorder.
I agree with Devon's reply, but I also wonder of you are looking for a gene burden test. That is, are there more rare variants in gene X (possibly damaging or not) in my cases compared to the number of rare variants in gene X in my controls (possibly damaging or not).
Take haemophilia A for example; a group of cases would have more rare variant in the Coagulation Factor VIII gene than controls would (it's a monogenic disease and almost all cases are explained by mutations in FVIII). It doesn't tell you which of the variants are pathogenic, but collectively, rare mutations in FVIII gene are more numerous in cases than controls.
Since you have a variant file (presumably vcf) of cases, and controls, you should have a look at plink/seq.
they have an awesome tutorial to work through some example data and you should to work towards the burden tests (under association tests)
Having said that, and with a nod to Devon's caution, you need to be sure the cases and controls are ethnically matched, and that the variants been called within the same exome (exomic?) regions, and sequenced using the same method, and have sufficient power to detect a significant enrichment, and boldly go where no man has gone before...
Your ethnic matching comment is really important. It can't be stressed enough how easily it is to shoot yourself in the foot if you don't match populations as closely as possible.
Hi rbangali
Thanks for your suggestion. I tried plink/seq on my vcf. But I am clear in interpreting the outcome the plinkseq.
Bellow is an example of plink association output
NM_000445 chr8:144990997..144991082 PLEC 5 BURDEN 1 0.2 7/12
Can you please help me intepreting it mainly the last field (7/12)?
7/12: 7 non-reference genotypes in the cases/12 non-reference genotypes in the controls.
BURDEN: is the test used
1: is the p-value
0.2: is the I value (it sort of tells you how significant the result could be, and it seems you will not be able to reach a significant value with this gene in this analysis)
Hi can someone help me with the interpretation of these results please:
LOCUS POS ALIAS NVAR TEST P I DESC
NM_000016 chr1:76190072..76229221 34,ACADM 64 BURDEN 1 0.2 313/147
I only have 12 cases and 28 controls - and I get (313/147) calculated from 64 variants? I'm not sure whether this number is supposed to be within the range of the number of cases and controls one has...
Also what is the usual threshold for I, P?
cheers
Log in to answer this question.
