Multi-allelic SNPS

[Jamee Berg]

Hi, Tanya,

Thank you so much for your amazing work on the FUMA website!

Can you confirm that FUMA handles multi-allelic SNPs by matching them to the LD matrix and will also drop palindomic SNPS, either because they are not represented on the LD matrix, or there is internal software that drops them?

For example, if I have these two rows in my sum stats:

REF ALT rsID
A G rs2758118
A T rs2758118

Would the second row be dropped?

If I have:

REF ALT rsID
A G rs2758118
A C rs2758118

Would the row that matches the LD matrix be retained for analysis?
I did not provide positions in my GWAS sum stats. 

Thank you!

Jamee

[Jamee Berg]

Hi, Tanya,

I re-ran my sum stats removing palindromic SNPs as well as indels. The results changed slightly. I noticed that FUMA retains rsIDs that are indels, so I am wondering if it is better to use the version that keeps the indels or not. If FUMA removes palindromic SNPs automatically, then the only difference between my two runs would be the indels. Do you think indels are informative?

Secondly, I would like to use the figures generated by FUMA for publication. Is think OK? If so, do I just use the citation on the website? Thank you!

[Tanya Phung]

Hi Jamee, 

There were many parts to your questions so I'll try to do my best to address them, but let me know if I had missed anything. 

First, the matching of SNPs in the user's input GWAS sumstat to the reference panel that FUMA uses depends on which information exists in the user's input GWAS sumstat (whether or not effect allele and/or non effect allele are specified). These decisions were made by the original FUMA  developer. But based on your information, I am assuming that you submitted to FUMA a file that has rsID, effect allele, and non effect allele. 
In the first step of processing the user's input GWAS sumstat, in this case (where rsID, effect allele, and non effect allele are present), what FUMA does is that it uses dbSNP version 146 to look up chromosome, position using the rsID. Because the effect allele and non effect allele are present in your file, FUMA checks if these alleles match regardless of effect allele or non effect allele. 

As an example, this is information in dbSNP v146 on FUMA: 
1       15274   rs62636497      A       G,T

The code was written such that only the first one would be considered. So when I submitted to SNP2GENE a file like this (toy example)
REF ALT rsID P
A G rs62636497 5e-10
A T rs62636497 5e-10

The first line  A G rs2758118 would be considered, and the second line is not.  Then, it checks that A matches with the reference and G is one of the alleles in the alternate options. So it will save the variant with the effect allele and non effect allele as in the input gwas sumstat. At the end of the processing, this is the variant that is kept: 

chr     bp      non_effect_allele       effect_allele   rsID    p
1       15274   A       G       rs62636497      5e-10

For your specific example of rs2758118,this rsID does not exist in dbSNP v146 that is stored on FUMA, so in this specific example this variant would not be considered. 

For your questions on indels, indels are included in the reference panel, according to the tutorial (https://fuma.ctglab.nl/tutorial#snp2gene): 

So if your indels match with the reference panel, then it will be kept. Whether or not they are informative to your specific analyses, I cannot answer that. 

For information on citation, please check the homepage of FUMA: 

Best,
Tanya

[Jamee Berg]

Hi, Tanya!

I was wondering if it is necessary to remove palindromic SNPs prior to running SNP2GENE. 

For my jobs, I have provided the following columns:
Chromosome
rsID
p-value
affect allele
non-effect allele
Beta
SE

My job IDs are:
748999
and
748949

Thank you so much for advising!

Jamee

[Tanya Phung]

Hi Jamee, 

I'm not sure if I understand how this question is different from your previous question. 

If your question is about how FUMA processes your variants based on the information you provided, you can check this in the log file. I checked for job 748999, and it says: 
2026-06-26 22:36:06,514 [INFO] __main__: Either chromosome or position column is missing. Look up from dbSNP will be performed to extract the missing information.

So this goes back to what I mentioned previously: 

what FUMA does is that it uses dbSNP version 146 to look up chromosome, position using the rsID. Because the effect allele and non effect allele are present in your file, FUMA checks if these alleles match regardless of effect allele or non effect allele. 

As an example, this is information in dbSNP v146 on FUMA: 
1       15274   rs62636497      A       G,T

The code was written such that only the first one would be considered. So when I submitted to SNP2GENE a file like this (toy example)
REF ALT rsID P
A G rs62636497 5e-10
A T rs62636497 5e-10

The first line  A G rs2758118 would be considered, and the second line is not.  Then, it checks that A matches with the reference and G is one of the alleles in the alternate options. So it will save the variant with the effect allele and non effect allele as in the input gwas sumstat. At the end of the processing, this is the variant that is kept: 

chr     bp      non_effect_allele       effect_allele   rsID    p
1       15274   A       G       rs62636497      5e-10

If your question is about in the case of ambiguity (for example 1       15274   rs62636497      A       G,T) whether or not these variants are trustworthy and/or whether the FUMA algorithm to match with dbSNP146 is reasonable, I think this is something you have to decide for your dataset. You could, for example, check, for the genome-wide significant variants in your file (variants where p<5e-8), are the majority of them not palindromic SNPs? For those that are, which ones are kept? If you need, in the SNP2GENE job submission, you could select the option to keep the input files. After the job is done, I can send you the processed file (that is used for downstream analyses) (please send an email) so you can check how your palindromic SNPs match with dbSNP146. 
You could also compare results with and without removing palindromic SNPs.

Best,
Tanya