Hi there!

I am not sure if GC and length are really the best measurements to pick. What about aligning the the similar genomes and then go for genetic diversity? Look for coding snps and pick all the genomes that have a possible functional divergence.

I hope this work by us in which we compared on the basis of SNP thru the agency of principal component analysis some genomes of mice strains could be of use for you:

http://www.la-press.com/a-novel-multi-scale-modeling-approach-to-infer-whole-genome-divergence-article-a3417-abstract

Hi,

first you should determine what is going to be your "taxonomic resolution" - whether you want to have wide groups, such as phyla, or narrow ones (genera, species, strains). Your strategy would depend on that.

When you want to stick to deep level, narrow groups, I think that Phillip Schiffer's answer is highly relevant.

If you want broader groups, you should use some synthetic measure of similarity, such as average (over whole genomes) similarity of protein sequences coded by the genomes. The expression for this could be like that:

S = ((sum of similarities/number of genes in genome1)+(sum of similarities/number of genes in genome2))/2

Alternatively, you could use megablast similarity combined with coverage, but you have to average it over the two genomes:

S = (similarity * coverage w/respect to genome1 + similarity * coverage w/respect to genome2)/2 where similarity is averaged over all hits and coverage is summed for all hits.

The reason for this approach is that for highly divergent genomes SNP number stops being usable metric. Basically, it works well up to ~3% divergence - a level of bacterial "species".

Another possibility would be to make two dendrograms, one based on the core genome and the second based on the entire genome with accessory genes included. Based on the dendrograms decide on what level of relatedness will provide you with a number of genomes for further analysis that you would consider to be reasonable given resources and time and choose genomes for further analysis that appear to be representative of key groupings. Comparing the phylogenies detected with core genomes to phylogenies with accessory genes included may help you to identify strains or isolates that have atypical accessory gene content, have higher than average horizontal gene transfer, or otherwise look interesting.

I would wonder why you would want to remove any genomes but if you do you need to I would look at how things actually happen in evolution.

For a given group of genomes that are (sometimes arbitrarily) grouped into a species there generally aren't huge differences in overall genome sequence (if they are actually grouped realistically) so there will be no overall differences in GC content. However, within a group, individual isolates may have picked up plasmids or integrated stretches of DNA that may or may not have a divergent GC content. Some call these islands and they can be very interesting, but you could probably rationally exclude genomes that have differences only in these islands. Unfortunately these islands can cause enough phenotypic variation that isolates may be grouped as a different species even though the genomes are almost identical otherwise.

The take home message is that for each "species" you need to choose a reference genome and then do sequential and group whole genome alignments to see how the genomes actually differ and then make rational guesses based on the guessed evolutionary relationships of the genomes ...... very labour intensive. Then exclude.

What you are really doing is trying to model the evolutionary history of every species in your study group ... not a easy task.

For many groups species is a fairly undefined concept, You might want to pretend it doesn't even exist and model every genome as a independent entity and see how it comes out.

Hi all,

Very interesting question, and nice posts.

Just a short comment @Glenn Soltes:

"For a given group of genomes that are (sometimes arbitrarily) grouped into a species there generally aren't huge differences in overall genome sequence (if they are actually grouped realistically) ..."

This may hold in general for %GC content, but variations in other measures, e.g. gene content, may be important and should be considered. Take for example E. coli strains, where several complete/near-complete genomes are publicly available: protein coding gene numbers vary from ~4200 to over 5000 ...

The way to choose a reference genome depends on the biological question asked and the resolution needed (cf. Marcin Golebiewski's comment). Taking into account that intergenic regions in prokaryotic genomes tend to be really short, I would rather rely on a measure related to coding regions/gene content in order to rationally pick a representative. For example, I would avoid choosing a strain if there was evidence that many of its genes were acquired by Horizontal Gene Transfer, unless of course HGT is relevant to my biological question.