FUMA code for obtaining LeadSNPs

[Carolina Makowski]

Hi Kyoko,

Firstly, thank you for creating and maintaining such an amazing web-based tool! 

I had a question about the code for obtaining LeadSNPs in FUMA.

I have several replication analyses I would like to do on many phenotypes - For these replication analyses, I only need the number of LeadSNPs for each of these phenotypes, without the in-depth functional follow-up that FUMA provides for my main analyses.

I have been using plink to try to replicate the number of LeadSNPs that FUMA provides but am finding that my plink results are always giving me a slightly large number of hits compared to what FUMA gives me.

Specifically, I have been using the default FUMA parameters of defining independent SNPs at r2=0.6, and lead SNPs at r2=0.1 with 250kb LD blocks.

To replicate this in plink, I tried to do a two-stage process of initial round of clumping with r2=0.6, then a second round with r2=0.1 on the SNPs obtained from stage 1, and as mentioned this is giving me a different number of hits compared to FUMA.

I suspect this may be because of the difference in the reference genome but would welcome your thoughts on this as well.

Is there a way I can obtain the code for FUMA's process of obtaining LeadSNPs, based on the parameters I used for my FUMA run? 

Thanks in advance for your help.

Carolina Makowski

[Kyoko Watanabe]

Hi Carolina,

I pre-computed pair-wise LD for 1KG Phase3 to speed up FUMA process.

I used PLINK with the following command

plink --bfile $pop/$pop.chr$i --maf 1e-4 \
	--r2 --ld-window-r2 0.05 \
	--ld-window 100000 \
	--out $pop/$pop.chr$

So the main difference is, --r2 flag by default computes squared of raw inter-variant allele count correlations.

While when you use --clump, it uses --r2 dprime which is based on maximum likelihood haplotype frequency estimates.

I never really compared how different they are but I recently discovered that it does make a difference when you clump with plink compared to FUMA.

One thing I noticed is that 250kb window is not to define LD block but it's just a threshold to merge LD blocks, in both steps of clumping in FUMA I use the same precomputed LD, so as shown above, I did not limit by the distance but the count of SNPs maximum 100000.

The full script is available at git hub

https://github.com/Kyoko-wtnb/FUMA-webapp

The script doing clumping is storage/scripts/getLD.py

(I'm sorry but this script is bit messy and might be hard to read...)

Also, pre-computed LD files are available from https://www.dropbox.com/sh/ysub3u91p73jkbs/AACLr_BTl2IZlfLWto7VgKXNa?dl=0

Let me know if you have any further question.

Best,

Kyoko

[Carolina Makowski]

Hi Kyoko,

Thanks for your response. That's very helpful.

I do have a question though because these pre-computed LD scores are for the 1k Phase 3 data only - I used the UKB Phase 2b reference panel in my FUMA analysis. Are the pre-computed LD scores for UKB available?

Thanks a lot 

Carolina

[Kyoko Watanabe]

Hi Carolina,

I am sorry for the slow response.

I cannot share the UKB reference panel at the moment since the access to the UKB is restricted by each application and under our application, I cannot share the reference panel (I can share GWAS sumstats but not the actual genotype data).

I am sorry about that and hope you understand.

Best,

Kyoko

[Andreas Schmidt]

Hello,

I found this old post, because I am about to clump some SNPs myself (as they do not reach the significance threshold needed for fuma clumping).

If I understand it correct, PLINK and FUMA clump in a different way (FUMA by r-squared and PLINK by D-prime)?

If this is true, I would get lead SNPs by plink with one method and indipendent significant SNPs in FUMA with another method...This does not seem to be a good strategy...Do you have any suggestions, how to do it?

Best,

Andreas

[Tanmay Ahmed]

Hi Kyoko,

The link for downloading  pre-computed LD files  are not working. Could you please share the updated link?

Best,

Saiful