IQTREE handling indels

[sammadge]

Hi, I had a quick query I wonder if you could help with.

I am using iqtree to analyse non-coding regions of closely related taxa. These have enough conservation to be alignable, however often contain gaps and indels, resulting in fairly gappy alignments. Often the indels are shared between taxa, and appear to be phylogentically informative. As I understand it iqtree treats gaps as missing characters. Is it therefore not taking into account these conserved indels in phylogenetic analysis? Is there a way to get it to treat indels as informative characters? If indels are simply treated as gaps is it better to trim away indels to improve alignment quality?

Thanks, any hep is appreciated!
Sam

[roblanf]

Hi Sam,

I moved this over to the discussion forum (issues are for problems with the software or the code, so keeping things separated helps us stay sane). Answers to your questions:

Is it therefore not taking into account these conserved indels in phylogenetic analysis?

Correct. It is ignoring the fact they are indels, and that they are potentially informative.

Is there a way to get it to treat indels as informative characters?

You can code them yourself. In other words, you can add an extra column to the alignment for each indel. Then you can use binary coding - e.g. a '1' for taxa that have the indel, and a '0' for taxa that don't. This is quite standard in phylogenetics.

If indels are simply treated as gaps is it better to trim away indels to improve alignment quality?

Not necessarily. If the alignment looks good where there is an indel (e.g. half the taxa have gaps, but the other half are well aligned) then there's likely plenty of useful phylogenetic signal in that region. Nevertheless, alignment trimming can be the crux of an issue in phylogenetics. My own opinion is that if different ways of trimming an aigment (all of them plausible) give you different answers, it means that you don't (or can't, yet) know which answer is best. Much like the treatment of outliers in statistics. The only honest thing to do is report both. Conversely, if you trim it a whole bunch of different ways and always get the same answer, you know that the answer isn't sensitive to trimming choices, and can report that.

I hope some of that helps,

Rob