Could I use a species tree to help build a reliable gene tree?

[Echodonut]

Hi!

I am trying to generate a gene tree, based on several genes in several species, all grouped by being in broadly one gene family.

Specifically, I am looking at about 30 heat shock proteins in C. elegans (hsp-16.1, hsp-16.48 etc), and all homologs that I found in a set of other species using Orthofinder (this includes paralogs within elegans, and approximately 25-40 orthologs for each other species totalling about 200 genes total).

What I have tried so far is to obtain protein sequences for all genes, feed them into multiple sequence alignment (MUSCLE), then forward the output into IQTREE2 and blindly let it run, default settings. This worked fine, although a few branches looked odd and so I tried to remove an individual gene and reran the analysis just to see the result, but found lots of branches got altered. This made me worry that the basis for the tree isn't as stable as it could be.

(my code, as run through the webUI: path_to_iqtree -s the_tree_+_CBG04606-CRE25952.aln-clustalw -m TEST -bb 1000 -alrt 1000)

I got the idea that it might be worthwhile to guide the tree-making process by incorporating a species tree, but I have no idea how to do that.

Would this be a valid method of improving my tree?
If yes, I would appreciate some direction on how to go about this, in IQTREE if possible.
If not, would there be a different way of achieving a reliable and robust gene tree?

[roblanf]

@Echodonut, I wouldn't advise this. For any reasonable alignment IQ-TREE should do a perfectly good job of estimating the ML tree without any attempts to guide towards a particular topology (e.g. a species tree).

Here are some things you probably should consider (with apologies if I didn't quite understand what you were doing):

1. Checking all of your alignments by eye (all of the inference assumes that the alignments are correct, so the better the alignments, the better the inference. Orthofinder makes plenty of mistakes, and alignment is a hard problem too...). You can then adjust the alignments by eye and/or by machine (e.g. by masking/trimming, etc). But your eyes are by far the best quality checker for alignments, and this will be the most important step in getting good inferences.

2. Running each gene tree individually

3a. Doing a number of replicates (say, 10) of each gene tree. The best tree is the one with the highest likelihood, but this will also tell you how stable the inference is.

3b. Doing bootstraps to assess uncertainty

4. Studying the gene trees. Any evidence of paralogy? Recombination? Etc.

5. Concatenating the gene trees which all seem to make sense (e.g. no paralogs or recombination) to get an estimate of the species tree

6. Using a method like ASTRAL to get an estimate of the species tree directly from the gene trees (here you can include paralogs).

Hope some of that helps,

Rob

[Echodonut]

Thanks the the extensive answer Rob!

I have been iterating on my trees by seeing which branches seem very anomalous, going back to the alignments, confirming if they are odd or normal (Orthofinder sure added a few genes into orthogroups which it probably shouldn't. Some have exact amino acid matches for only 2 positions!), and rerunning the tree without those genes. And I am running multiple copies of my trees now as you suggested, with much higher bootstrap and numstop values. I think when these have run, I should be able to make a safe call on what is the most reliable tree for my gene set!