In the prerecorded videos Ran suggested using a 3MB window around the lead variant, but I’ve also seen pipelines use 1Mb or even 500kb. How should we think about what window size is best to use?
Replying to “In the prerecorded videos Ran suggested using a 3M…”:
The bigger the region the more computationally expensive the fine-mapping becomes but the more likely you are to capture the true causal variant
Replying to “In the prerecorded videos Ran suggested using a 3M…”:
Great question, there are some limited investigations into how to set the regions. Looking at the LD is a good way to do it. You don’t want to leave out variants in non-trivial LD with the lead variant. In some regions, if the LD is low, you can use a smaller window, but some regions have extensive LD, you may have to use a big region. Usually computational constraint is the determining factor, secondarily, I usually wouldn’t use 3Mb if I run ABF because ABF has the single-causal-variant assumption, which only picks up 1 causal variant in the region. If the region is big, it may contain multiple causal variants that ABF will miss.
Replying to “In the prerecorded videos Ran suggested using a 3M…”:
Thank you that’s very helpful!