Dear all,
I have a collection of data for a continuous human trait, along with a genome-wide genotypic information for the same individuals. For the continuous trait, most individuals are centered around the mean (leptokurtic distribution). When performing the association study, is it more useful (more powerful) to use the individuals at the tails of the distribution (those phenotypically extreme) or the full distribution?
thanks
You will typically get better statistical power by using all the data, however, using the tails is interesting as it may reveal loci that contribute to large changes in phenotype.
Let's say if you target major loci contributing to the phenotype you are interested in then the extreme class of phenotype might gave you better results but still you need descent number of individuals in the analysis. On the other hand, when you consider all then you get all (therotically) loci beind the phenotype.
Vincenzo and Rajiv: thanks, very clearly explained. Now I can see it. My problem was related to the amount of money that i had to invest: seems more acost-effcient to go for the low-hanging-fruit (big differences between the tails) than trying to identify all possible loci that contribute, giventhat theyir effect is going to be low and consequently I would need to genotype an unaffordable amount of samples. Do you agree?
Robert: thanks, I was so blinkered into looking at the differeces... in fact the (or at least another) interesting point is finding out what keeps them so similar. Good point.
Santos: I would agree that taking extremes is a cost-effective strategy. There are many projects I have been a part of that use this scenario. However, be advised that you will likely uncover only a subset of loci associated with your trait and likely only those with large effect. You may want to perform a power analysis of what sample size you would need to observe given effect size. Do you have enough individuals in the extremes to provide statistically significant results?
A the moment I have available about 500-600 people, from which I could build two tails of 40 people each. I keep on sampling, but my budget is modest, i.e. I could not afford to genotype thousands of people. I must add that genotyping for association alone is not my single goal, I plan to haplotype the data, perhaps impute some other data, and do some population-genetics tests.
Again, depends on expected effect size. Imagine your effect size is represented as the slope of a line (beta). For small effect size, ie beta near zero, you need sufficient sample size to be confident that observed departure from 0 is statistically significant. For larger effect size you would need fewer samples as deviation from 0 is easier to statistically support.
You can do both? A ‘continuous’ analysis, where you have people varying continuously between the two extremes; and a ‘categorical’ analysis, where use a subset of the data you collected by only analysing the differences between the top and bottom performing people for that trait.
How to define a continuous phenotype. If I have total 15 phenotypes of bacteria, how to perform continuous phenotype analysis?
- Badges
- Science topic