Hi Ajit, thank you very much for your interesting overview about GBS :) Unfortunately I am not working with GBS data and my questioning was more about genetic diversity metrics that could be use in a specific dataset of only ~100 SNP obtained through classical SNP-typing (mass-spectrometry in my case).

e.g are classical He/Ho, polymorphism rate or private alleles sufficient, or should one use complementary approach to measure/compare genetic diversity ? And which one ?

Of course, it seems totally relevant to combine as much information as possible, from various measures just like you said. But I am afraid I am missing another kind of approach aside classical genetic diversity analyses.

Thanks again :)

Chrystelle,

for SNP data I am currently using the PopGenome package in R (http://popgenome.weebly.com/). It allows you to estimate several diversity statistics within and between populations. Now to answer your question, I would focus in nucleotide diversity, number of polymorphic sites, number of private alleles.

Hope this can be of any help,

Thank you Manolo and Ajit for your response and remark :)

Manolo, unfortunately I have a very restricted access to sequence data (all I have are consensus sequences resulting from pooled RAD-seq data from a large geographic area, therefore I do not have any individual haplotype information). However, it is reassuring that you also suggest polymorphism rate and private alleles as relevant measures for diversity with a small number of SNP :)

All the best !

Chrys

using 100 SNPs (assuming a reasonable minor allele frequency) is not such a difference in resolution compared to classical AFLP studies. 

gene diversity is nothing else than the expected heterozygosity (He). If you have an average gene diversity of c. 0.15 - 0.20 (which is the chance that if you take two random alleles from your sample they are different), you're in the typical range for this type of population genetic studies. 

In the early days, many allozyme studies had similar diversity ranges (but fewer loci). So it's pretty OK to use He, compare it to Ho, calculate allelic richness, and so on. The more loci, the smaller the error on the estimate of genome-wide diversity and so on. So more loci allow you to discern small effects from noise better. 

Basically, all classical population genetics is based on detecting deviations from Hardy-Weinberg expectations (F-statistics at different hierarchical levels) to test for evolution. (the HW theorem assumes no evolution, hence it is used as a null model). Any deviation from HW is caused by evolution, and the trick is to disentangle the three (four) major drivers of evolution: drift, gene flow (and subdivision), selection (which is different per locus due to recombination) and mutation (which is in itself typically negligible compared to the three other).

More importantly, what questions are you trying to answer? Do you require individual-based analyses (assignments, clustering, parentage, ...), are you satisfied with estimates at the population level (summary statistics), etc.

As a reviewer, I'd rather read a manuscript with a good sampling design for the question asked, crisp statistical analyses, clear knowledge on the assumptions of each analysis (the biggest issue in population genetics if you ask me), and using only 100 SNPs, than a study that uses 100 000 SNPs but has no clue how to really answer the question (other than "we were interested in ..."), other than by blindly copying what others have done in somewhat similar studies.  

Thank you very much Joachim, for your very enlightening answer !

In my case, considering I am working with non-model species; being able to develop a correct (i.e relevant and reliable) number of SNP led to questions I think I would have perhaps neglected otherwise (e.g with classical approach like SSR based studies).

For now my He values typically range from 0.25 to 0.35 depending on the species, and in most cases it seems like my modest number of loci is sufficient to detect substantial population structuration through summary statistics like Fst. I was a bit more concerned about the relevance of comparing, for example, genetic diversity between two different sampling stations since, if taking mean per-locus allelic richness Ar as an example; Ar will typically be restricted to the narrow range of 1 to 2 (considering most of SNP markers being biallelic). 

From now on, I need to find the best fitted way to rigorously compare degree of genetic diversity between different grouping of individuals with my markers.

Again, thank you very much :-)

Best regards,

Chrys