Hello,
I am involved in analysing the single nucleotide polymorphism of a certain gene from 4 different geographical populations and so was wondering what percentage/threshold level that I should use to say that a position is really polymorphic for that gene? Please advice or give some references that I can use for my study...
Appreciate your time and help.
Thanks
With low numbers of samples and a high testing reliability you might say that a SNP is truly polymorphic if two variant alleles had been observed in two different individuals. One other way that the likelihood could be estimated is by carrying out a chi2 test comparing your observed allele counts with an expected count of zero.
The classical SNP definition required a MAF of 5%, but that's been relaxed considerably now that true variants that have very low frequences have been discovered. I wouldn't trust a polymorphic definition based on a single typing method, but if you found a SNP with similar MAF using two different genotyping methods (and observed in at least 2 people), I could believe that.
Usually, the frequency of minor allele should be more than 3-5% in a population. In this case, the SNP is really polymorphic.
Hi, I want to adress a similar but different question within this topic. I have a study population (n=50 subjects) and found n=19 SNPs in the candidate gene. Now I would like to determine a kind of association between my observed genotype and clinical phenotype.However, I do not have data on controls as we did not include these in the study for genotyping. Therefore, I was wondering I it is allowable/justified to use a representative population genotype data from hapmap or 1000 genomes and compare these to our genotype data. In addition, SNP that deviate from controls I would like use to determine associations. Any thought on this?
If the population prevalence of the phenotype of interest is low, then it should be fine to use 1000G or HapMap. If the phenotype has a moderate-high prevalence (say >5%), then you'll need to use a population that has been enriched to exclude the phenotype of interest. At worst, you will fail to discover associations when they exist by using non-enriched control populations.
Bear in mind that there may also be haplotype associations that aren't represented by the SNP associations alone. If you're looking at a single candidate gene, I'd recommend doing a haplotype block analysis to see if there are any SNPs in high LD that could be used for doing haplotype-based association testing.
Can anyone suggest me an online windows based assembly software for consensus contigs and then analysis through SNP?
Dear David,
Thanks for your answers. However, I was curious to how you came up with such a threshold? Is this a general rule of tumb? Do you perhaps have some references on this?
I wouldn't really call any of my values a threshold as such -- there are a lot of greyzones when you get into the details of diagnostic work. Regardless, most of my "enriched" ideas are from my own research looking at genetic markers for T1D (chapter 5 of my PhD thesis). You just run into problems with determining the utility of your classifier.
With a prevalence of 5% in a non-enriched population, the maximum possible specificity will be 95%. While that sounds great, it plays havoc on the positive predictive value of tests, particularly when a phenotype is hidden (e.g. something that only manifests later in life unless tested prior to that).
So, to be short, in my case the prevalence of my phenotype is below 5% and this would "justify" a simple comparison with the general population. However, any significant deviations may only provide a tiny clue and must not be extrapolated? Is this a correct interpretation of your words? In addition, I do like to read chapter 5. In this chapter, do you also use such an approach?
Any association-testing is a hypothesis-generating approach; as long as you're aware of this (and report the limitations), I consider it reasonable to extrapolate from large non-enriched control studies to the general population. However, any reasonable study should include validation of your hypotheses, so you're going to need some proper controls (and other cases) eventually anyway.
I dealt with the validation issue in my research by concentrating on the method, rather than the results. It's nice to have results that might actually be useful, but the thing I'm presenting is a general approach for how to do it in other populations / groups. I did a bit of internal validation via population sub-sampling and independent testing to confirm the associations (a bit easier with the WTCCC populations, which have thousands of individuals), but wouldn't want others to use my markers without seeing the results of validation in another population first.
Do you think it would be more informative to look at it in another way, eg skip thet snp freq comparison and look at eqtls?
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication