Fine-mapping regarding 5 SNPs

[jacksondu76]

Dear all:
First of all, thanks for the great package!
I just have a naive question since I am pretty new to fine-mapping process. For my research objective, I plan to fine-map 5 particular variants with CHR, BP, A1, A2, EAF, BETA, and SE. I know that PolyFun is advanced in handling in fine-mapping GWAS summary statitistics. I just wonder in my case (fine map only 5 SNPs), is the PolyFun tool and subsequent SuSiE going to produce reliable results? Or if it is mandatory to have the summary statistics for the analysis pipeline?

I would really appreciate your insights.
Best,
Jackson

[jacksondu76]

Sorry to mention that my 5 SNPs are independent genome-wide significant SNPs from my GWAS summary statistics.
I also want to ask another question. I noticed on the Wiki, step 2. Using and creating functional annotations,we were to add functional annotations of our SNPs, and also compute LD-scores for annotation. However, I noticed in the step 3. Functionally informed fine mapping with finemapper, it did not utilize the functional annotation but only uses SNPVAR. I wonder what the step 2:creating functional annotations is intended to do in this pipeline analysis; and if I utilize UKBB genotyped data per chromosome，there is no need to compute LD scores in step 2?
Again, thanks so much for your help and look forward to your reply!

[omerwe]

@jacksondu76 these are deep questions/ I have limited bandwidth to answer deep questions but briefly:

1. You need genome-wide summary statistics for reliable fine-mapping. Five variants isn't enough
2. SNPVAR is derived from the functional annotations, hence you need functional annotations, LD scores, etc.

[jacksondu76]

Thank you so much for your reply! I will use summary statistics for the analysis. I just have two very specific questions.

1. In the wki step 3, you performed Fine-mapping with SuSiE, using chr1 genotypes from a plink file, while specifying --start 46000001 \ --end 49000001 . In my case, one of my SNPs is located in chr2, BP=27730940. In this case, what are your suggestions to set up the start and end parameter? Is it reasonable to set up start and end position as 27730940 ±1 Mb?

2. You mentioned SNPVAR is derived from the functional annotations hence I need functional annotations and LD scores: I wonder if I used precomputed prior causal probabilities of 19 million imputed UK Biobank SNPs with MAF>0.1%, based on a meta-analysis of 15 UK Biobank traits, then there is no need to calculate LD scores and functional annotations on my own? In addition, 17283 out of 727352 SNPs were missing annotation information, when I performed
   python extract_snpvar.py --allow-missing --sumstats pdff_data/chr2.finemap_sumstats.txt --out output/chr2_with_var.txt.
   I wonder if this level of missingness could influence subsequent analysis, or do you suggest that I should adopt the approach 3: computing prior causal probabilities non-parametrically?

I would really appreicate your suggestions, and thanks very much.

Best,
Jackson

[omerwe]

1. in the paper we used a windows of size 1.5Mb on each side, but 1Mb is probably fine for most use cases

2a. Using precomputed prior causal probabilities is a compromise. It might work well if your traits are sufficiently similar to the 15 traits used for computing these, but I can't guarantee anything...

2b. Missingness is problematic -- if the true causal SNPs are some of the ones that are missed, you'll miss them... Sometimes there's no choice, but this is a limitation you'll need to keep in mind and deal with somehow...

In general it's probably always better to compute the prior causal probabilities yourself if you have the data, but it's a long tedious process. It's your call...