100 genome phylogeny

[mdk308]

Hello, I am trying to create a phylogeny with 102 different strains of Neisseria. Normally I would just crate a SNP table, derive a multi-fasta from the SNP table (i.e. create a SNP profile fasta for all SNPs within any given strain), algin those SNP profiles, and use a tree program to draw a phylogeny.

The issue here is that I have a lot of genomes and also that the genomes are pretty divergent (SNP counts ranging from 2k to ~30k) so the total list of characters in each fasta comes out to about 180k. Apparently aligning 102 fasta files that each contain 180,000 bases is too much for any of the tools that I am currently using.

So can anyone suggest an alternate method? From looking at other posts, it looks like I oculd try to align Mugsy and use that to create a core genome, which I would then use as the reference to call SNPs from (the theory here being that I would have a lot fewer SNPs if I was calling form the core genome as opposed to the reference that I'm currently using). My only hesitation is that I'm not entirely sure how much this will cut down on the SNPs. Any other ideas? Perhaps it would make more sense to just try to filter out SNPs form the current list but I couldn't think of a good methodology for this.

Thanks.

[themerlin]

If you have a SNP table of only core SNPs, they should all be the same length and you shouldn't have to do any subsequent alignment; meaning that when you convert your SNP table to a multi-fasta, all sequences should already be the same length and you can put directly into a phylogenetic program. Maybe I'm missing something, but the number of SNPs in your sample shouldn't matter.

Jason

[bckirkup]

Another strategy

The advice Jason gives is good, as far as it goes. He certainly knows quite a bit about the topic, including SNPs. However, I would not recommend a SNP-based phylogeny in general; though it works well for some closely related and clonal organisms. Neisseria seems to be ok; at least within a species; but as you go beyond the borders of the genus you would encounter real trouble.

As an aside, you should never have to actually align SNP profiles per se. Instead, they should be 'pre-aligned' unambiguously as they are found.

Anyway, in my work with groups of genomes, they generally diverged so much that synteny was complex and horizontal gene transfer common. SNPs were not going to be much use. The basic strategy has been to use something like OrthoMCL to find the core single copy genes; align them as families, and either build trees or build a phylogenetic network.

There are problems with this strategy. The set of core single copy genes is often very few; alignment is rather messy. So much information is lost during the process that resolution is poor at the tips of the tree.

An alternate strategy, for which I have some scripts, starts with the same basic principle but only uses those first genes to do the deepest branches. Then the tree is bisected and OrthoMCL is reiterated for the subtrees. The core genomes on each subtree are larger and create the next level in the tree. This process continues iteratively through the tree, with the subtrees becoming smaller - but the core genomes in each becoming much larger, more informative and less ambiguous. Finally, at the end, the tips are full resolved with most of the genome available and aligned.

Keeping track of the process also produces a map of when gene families appear on the tree and can be used to generate genome-wide information about gene content.

If you want the scripts or want to talk about phylogeny, just let me know.

[nickloman]

Given the way you produce the multi FASTA file, isn't it already an alignment? You should be able to use phylogenetic software directly. FastTree is a good option. However I suspect your tree won't make much sense because Neisseria are so recombinogenic. You could try removing blocks of SNPs suggestive of having origins from horizontal gene transfer.

[mdk308]

Thank you all for the feedback!

Yes, the multi-fasta is essentially aligned to begin with. The issue was that I did not have a placeholder character for indels so every fasta varied slightly with respect to total characters. Easy enough to fix. I didn't realize that I could just use the fastas as direct inputs for tree programs.

The larger question I was asking (which in retrospect wasn't too clear) was really whether or not this method makes sense given the nature of the Neisseria genome, and it sounds like maybe it doesn't. nickloman, i thought about pulling out the SNPs that appear to be horizontally transferred but I wasn't really sure of an automated way to do this other than to create a core genome. Is there an easy way to do this?

bckirkup, that sounds like a really sensible approach. I would be very interested in getting those scripts, if possible. I'll shoot you a message.

Thanks again!