3
$\begingroup$

Genetics datasets contain measurements for millions of single nucleotide polymorphisms (SNPs). Some (usually small) percentage of these values are of low confidence, and are labeled as missing values. It is common to impute these missing values using statistical relationships within the high-confidence samples as well as relationships mined from public datasets. This makes sense, to get an estimate of what that small percentage of missing data likely is in reality.

Sometimes, analyses are performed that impute up to a much larger number of SNPs - for example, a genotyping platform might measure 2.5 million SNPs, but imputation is performed to get a larger sample of 6 million SNPs. My question is, what value do the extra SNPs that were never even measured have from an analysis point of view? When performing feature selection or predictive analysis, it would seem that the imputed SNPs are really just encoding statistical relationships that are present in the dataset to begin with. What does it mean to find a highly associated or predictive SNP in the imputed set?

$\endgroup$
3
$\begingroup$

In GWAS we are interested in understanding which SNP has a causal influence on a specific phenotype. At the moment large scale studies are carried out by genotyping arrays. For each SNP we put on the array, we measure the alleles present in the patient.

The cost of genotyping obviously depends on how many SNPs we would like to measure. So the problem when designing a genotyping array boils down to: What SNPs do we put on such an array so that we get most information out of it.

To answer that question it is important to realize that SNP alleles are not independent. SNPs that lie in the same haploblock are highly correlated. So measuring one of these SNPs is often sufficient to predict the alleles of other SNPs that are in so called linkage disequilibrium. When designing a genotyping array, we probably do not want to measure all of those highly correlated SNPs, because we gain very little information. Instead may want to measure just few of them and then impute (predict) the alleles of the correlated ones from population study data.

So back to the GWAS analysis. From the SNPs on the designed genotyping array we now predict SNPs that are associated with the phenotype. Say after our statistical analysis we found an associated SNP. But from the design of our array we know that there may be highly correlated SNPs that we did not measure. All of them will also be statistically associated. If we do not impute, we may come to the wrong conclusion that we already found the true causal SNP. In fact, there may be many and finding the correct is often very difficult.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.