1

In many association studies (e.g., GWAS), a large number of Linear Regression models are fitted. Then, a strategy to account for the Multiple Testing issue is adopted (e.g., Bonferroni). That being said, it is clear now that it's not that easy, and a large number of false positive results are retained.

I am now trying to understand whether the use of Bayesian Regression can help solve this problem. I am just now reading some material on Bayesian Statistics, so maybe I am totally wrong. I read that if you use Bayesian models, you don't have to worry about this issue. First: is this actually true? Second: what does it mean to perform an association study using Bayesian models? Does it mean fitting multiple Bayesian regression models?

wrong_path
  • 607
  • 6
  • 20
  • You have to worry about how to select your prior in Bayesian analysis. When you have multiple hypotheses, you have to specify a joint prior on the parameters and hypotheses. The problem does not go away at all, you just have to deal with in a different way. Simple shrinkage is only appropriate under a very restrictive prior assumption about the science, one that does not apply to GWAS, eg. – BigBendRegion Jul 30 '21 at 19:50

1 Answers1

1

(Answering because cannot comment).

Here are a few links that might be helpful.

I don't have experience with GWAS studies but in general, bayesian methods do not need to be corrected for multiple comparisons. It is my understanding that this is due in part to the use of an appropriate prior - from Andrew Gelman's blog:

"...with normal data and a normal prior centered at 0, the Bayesian interval is always more likely to include zero, compared to the classical interval; hence we can say that Bayesian inference is more conservative, in being less likely to result in claims with confidence."

As far as an "association study", that sounds to me like bayesian linear regression...

jpalm
  • 81
  • 1