You will save computational space!

Thanks Gezahegn. That is true but I wonder if it is reasonable to make comparisons between different populations using different numbers of SNPs If some populations have higher proportions of monomorphic loci?

Hi Elizabeth,

I think it is reasonable to remove a set of loci if they are monomorphic in all the populations (in your case in all the regions sampled). The reason being they will not be informative. But if in some populations the density of monomorphic loci is high (given that same set of SNPs analysed), it might signify low genetic diversity in those populations.

Thanks for your answer. I would agree, perhaps the key thing is that you need to use the same set of SNPs? But what if you use different SNPs for different populations / regions - then is the comparison more robust if you remove the monomorphic loci?

Hi Elizabeth,

I'm also thinking about this, I reckon that after filtering for monomorphic SNPs in the entire database, it makes sense to me to report the unique monomorphic SNPs per region/population/etc. For downstream analysis, I tried removing unique monom per region and only keeping the common polymorphic SNPs, but it feels like I'm "forcing" something here and losing valuable differences. Still not sure as I can't find much info about this.

Best,

Laura

In your case, it is definitely a mistake. The things you are describing are probably very useful semispecific alleles or loci in which regions are differentiated. A different thing would be if there is no differentiation across all regions, in the whole dataset. Take a look at this: https://www.biostars.org/p/80014/

People usually suggest excluding these monomorphic loci from STRUCTURE analysis because there is no differentiation or any variation in these loci and they bring no information on ancestry. This means that using a dataset with some monomorphic loci versus excluding them would lead to the same results (or with negligible differences?). I have checked a hypothetical example, and I am not so sure about this... :

7 individuals, 3 pre-defined populations, and 2 dominant loci in which they are fully differentiated (an alternative was +8 monomorphic loci).

In the Evanno tables, you can see the difference in the delta values...

Of course, I forgot there will still be some differences between results since it is a stochastic process, and it is the relative pattern which is important, but I still think it is not very justified in absolute terms. (it is multilocus data as a whole, not the locus-by-locus thing).