# Is there any difference between SNPs 'AG' and 'GA' in association analyses?

I am using plink to check the association between a phenotype and the SNPs of a gene. plink says the phenotype is significantly associated with a SNP on that gene, and when I check the SNP alleles, for 6 out of 49 samples, the nucleotides are 'GA', and for the rest, they are 'AG', so only the order is different. I am not a biologist, but my basic genetics knowledge says that those two are semantically the same. However, obviously, plink does not agree with me and it associates this SNP with a small p-value since 'GA' samples have a higher value for the phenotype than 'AG' samples. I am very new to SNP analysis, any help is appreciated.

Could you please show us the context in which this appears, as you seem to be interpreting this differently to Devon.

If it's appearing as you say, GA and AG, then Devon is right, this usually means that the sequence goes ###AG### or ###GA###, which are two very different sequences.

If, however, as you're implying, the sequences are actually ###G### or ###A###, then semantically, the two ways of writing it have the same outcome. Normally, however, this is written as A/G or G/A.

Alternatively, you're looking at genotypes, which would usually appear as A|G and G|A. In this case, both individuals are heterozygotes, and in terms of the individual SNP they are the same. However the side of the pipe is relevant as this tells you which homologous chromosome it is on. This allows you to calculate LD with respect to neighbouring SNPs, and to determine the actual sequences of the proteins in that individual.

• Thanks for the reply. I am looking at genotypes. I realized that I had a bug while retrieving the genotypes, so what plink thinks as being different was not in fact GA vs. AG (each of which would really refer to heterozygous individuals as I replied to Devon), but it was GA and AA, where one is homozygous and the other is heterozygous. So, it is totally normal that those alleles are different for SNP association analysis. Aug 22 '17 at 21:15
• "However the side of the pipe is relevant as this tells you which homologous chromosome it is on. This allows you to calculate LD with respect to neighbouring SNPs, and to determine the actual sequences of the proteins in that individual." -- Could you please explain this a little further? Aug 22 '17 at 21:55
• If you have two variants, separated by, say 50, bp and they both fall in one gene. The genotype in one individual is G|A for the first variant and C|A for the second. That means that one of their genes contains a G and a C, the other gene contains an A and an A. This might be relevant in various ways. In a population study, you might just be generally interested to know whether G...C and A...A are more common than G...A and A...C. This would also matter in a phenotype association study, as only one may be causative, but if they always turn up together then how would you know which? Aug 23 '17 at 7:39
• (ran out of space!) It could also be that the phenotype only occurs when both variants occur, or is perhaps alleviated by the pairing. If both those variants affect amino acids, and those two amino acids are in contact, mutating one only could disrupt the connection, but mutating both the connection is maintained. Or if they're functionally redundant amino acids, one is able to compensate for the other, so when only one occurs, there's no phenotype, but when both do there is a phenotype. Aug 23 '17 at 7:45
• There's a pair of courses here on introduction to genetic variation, which explains a lot about things like LD, effect on proteins etc: ebi.ac.uk/training/online/course/… ebi.ac.uk/training/online/course/… Aug 23 '17 at 7:46

Is "user5054" the same as "user5504"? No? Exactly. Not only does order matter, it's incredibly vitally important. AG and GA are completely and totally different from each other. If this is in a coding region, then the resulting amino acid is undoubtedly changed (fun fact, the only exception is TAG and TGA). If it's at a splice site then splicing is likely altered (AG is part of a splice acceptor site). If it's at the binding site of something then it wouldn't be surprised if it didn't bind any more.

• I think "user5054" vs. "user5504" example is misleading. My example would be that the sets {1,2} or {2,1} mean exactly the same thing, while if those represent permutations, they are different. So it depends on the case. In the SNP case, each of 'G' or 'A' in 'GA' or 'AG' are the nucleotides in the one of the two strands of DNA, right? So, both 'GA' and 'AG' are heterozygous alleles for the same SNP.Then what I understand is that the ordering of the strands matter,because maybe the transcribed strand's nucleotide is written first while representing the SNP?I think your reply needs more detail. Aug 22 '17 at 7:36
• Is this actually a SNP or is it an InDel? If it's a SNP then you're correct that it doesn't matter (assuming no phasing information). My presumption is that this is not a SNP, but rather an InDel. Aug 22 '17 at 7:50
• Yeah, this is a SNP analysis. I should have been clearer when I asked the question. Please see my comment to Emily_Ensembl above. The statement in my previous comment to your reply saying "...the ordering of the strands matter, because maybe the transcribed strand's nucleotide is written first while representing the SNP?" was completely out of confusion due to being new to genetic analyses. Apparently, Each of G or A in "GA" or "AG" do not refer to different strands of DNA, but refer to the different chromosomes of an individual. Aug 22 '17 at 21:20

Most association analyses are carried out at a single SNP level, so AG and GA are likely to indicate a heterozygous genotype at a particular location. However, the precise notation matters.

As @Emily_Ensembl has alluded to, for VCF files, A/G indicates unphased SNPs (order unknown, and shouldn't be considered in analyses), whereas A|G indicates phased SNPs (order matters).

It is possible to carry out association analyses by combining multiple variants into a single haplotype, in which case the order would also matter. It's unlikely that this is the case, because such haplotype variant analyses are substantially more complicated to carry out, and will typically be paired with information about SNP linkage (statistic D´ or r²).