Is neighbor joining the best approach to look at clustering pattern with population genetic data?

Question

[su7880]

Hello,

I am working on a big SNP data set from GBS with over 300 individuals from 34 populations. 34 populations compose three closely related species. I tried various assignment tests to see the pop structure but still wanted to see clustering pattern with different approach. Unfortunately, I am not an expert of tree building. For a starter, I am not sure neighbor joining will give me informative inference on relationships among species and populations. Also, there are many heterozygote individuals for many loci since I am using SNPdata set. Which software takes account for ambiguity codes if I do neighbor joining analysis with my SNP data?

Any kind of answers will be very much appreciated.

Thanks in advance.

[reza.jabal]

Hi, It might worth running PCA first on your data. Random Forests are roustabout classifiers, and having reduced the number of your features (SNPs) to manageable size you can build a nice model.

Answer 1

[su7880]

Thanks for the kind answer.

I already tried PCA and various assignments tests like fastStructure, DAPC. However, I still want to look at what NJ does with the data that I have. Any more suggestions?

Answer 2

[Brice Sarver]

With that kind of data, you'll want to use a distance-based approach like NJ, UPGMA, etc. More sophisticated phylogenetic methods are unlikely to get you an answer in a reasonable amount of time, especially if you're just looking at your data. You can also use a non-phylogenetic method, like hierarchical clustering, if you just want to explore your dataset.

[reza.jabal]

Hi Brice, out of interest I am wondering why you are not recommending random forests? Are they not outperform other models?

[Brice Sarver]

RF might be useful here. I just honed in on NJ, species/population data, and tree. On a second read, perhaps you weren't explicitly talking about using phylogenetic approaches and a dendrogram, as opposed to a phylogeny, will suffice for what you want.

Answer 3

[su7880]

Thank you all. I am sorry what is RF? Can you give me some more details for that? Also, do you know software that might actually take account IUPAC ambiguity codes for phylogenetic inferences? I just saw an argument that MrBayes might or might not use the IUPAC ambiguity codes. Is RAxML a good? What do you all use for this kind of question?

Many thanks in advance.