This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Substitution model selection with IQ-TREE (ModelFinder)

Hello, everyone. I need advice on substitution model selection with IQ-TREE (ModelFinder).

I am conducting phylogenetic analysis on plant's amino acid sequences. Checking my results of the .best_model.nex file, I noticed that some models for the sequences are Q.MAMMAL and Q.BIRD, the substitution rate of which seems to be calculated based on the animals data. I used -m MFP for model selection.

Does anyone have the same experiences? And should I do the analysis again after restricting models used for the model selection?

I am glad to hear any advice or comments.

selction iq-tree modelfinder model substitution
• 679 views • link
updated by Mensur Dlakic Mensur Dlakic • written by D • D

2 answers

Hi,

I think you could try:

  1. Restrict the search space: Use -mset (to limit to general amino acid models) and compare the outcomes with the previously obtained results looking for the best biological significance.
  2. You can compare the topology and bootstrap between the 'Animal' model and the restricted model

Your also could use ModelTest-NG (https://github.com/ddarriba/modeltest) and compare the chossen model too.

I hope you find this useful

• 0 views • link
written by Leonardo • Leonardo

Thanks for your advice! I will try -mset and compare the outcomes. I also used ModelTest-NG, and then most of the results were JTT model. That's why I feel how I should do.

• 0 views • link
written by D • D

ModelFinder will test many models, and it should make no difference that some of them are "animal" models. What matters is the model that was selected as best based on a high likelihood score. Many proteins are highly conserved across eukaryotes, making even animal models applicable to plants.

As IQ-TREE has no knowledge of your sequences' background, it will select the model that fits them best. I don't know if there is a clear rationale where a general amino-acid model would be expected to be better for modeling plants than an animal model. What exactly was the best model selected - I usually look for those in a file that ends in .iqtree - and why would an animal model be a problem even if selected as best?

• 0 views • link
written by Mensur Dlakic Mensur Dlakic

Thank Dr. Dlakic for your comments!

List of best-fit models per partition:

  ID  Model                  LogL         AIC      w-AIC        AICc     w-AICc         BIC      w-BIC
   1  Q.PLANT+R5       -10268.517   20555.035 + 6.95e-310   20555.603 + 6.95e-310   20589.144 + 6.95e-310
   2  Q.MAMMAL+I+G4    -22638.665   45283.331 + 6.95e-310   45283.370 + 6.95e-310   45296.605 + 6.95e-310
   3  Q.PLANT+G4        -8627.735   17259.469 + 6.95e-310   17259.517 + 6.95e-310   17266.575 + 6.95e-310
   4  Q.MAMMAL+I+G4    -10711.367   21428.735 + 6.95e-310   21428.822 + 6.95e-310   21439.639 + 6.95e-310
   5  Q.PLANT+R4        -9976.281   19966.561 + 6.95e-310   19966.905 + 6.95e-310   19993.239 + 6.95e-310
   6  Q.BIRD+I+R4      -25223.685   50463.370 + 6.95e-310   50463.621 + 6.95e-310   50498.316 + 6.95e-310
   7  Q.PLANT+I+G4      -8152.883   16311.767 + 6.95e-310   16311.860 + 6.95e-310   16322.472 + 6.95e-310
   8  Q.MAMMAL+R3      -12631.473   25272.947 + 6.95e-310   25273.080 + 6.95e-310   25293.559 + 6.95e-310
   9  Q.MAMMAL+R4      -15962.470   31938.940 + 6.95e-310   31939.226 + 6.95e-310   31966.880 + 6.95e-310
  10  Q.PLANT+R4        -9829.733   19673.467 + 6.95e-310   19673.896 + 6.95e-310   19698.630 + 6.95e-310

I show 10 examples from the list of best-fit models per partition from .iqtree file. My AA sequences are from plants but Q.MAMMAL or Q.BIRD were selected as best. Seeing this results, I wonder if Q.MAMMAL or Q.BIRD are most appropriate even though plants and animals may have different amino acid composition and evolutionary constraints on their proteins. As I am new in phylogenetics, I may have lack of knowledge and say incorrect things.

• 0 views • link
written by D • D

I am only guessing here from the output, because you didn't give us all the information. It appears that you have partitioned your alignment, and from individual partition scores I would guess that your partitions are not very large. Maybe individual genes in a concatenated alignment? If so, I would say that this results is probably legitimate. Based purely on numbers above, it seems that the shortest partitions are all Q.PLANT.

Do you have a reason to believe that your partitions evolve differently? If not, you may want to determine the best model for the whole alignment, and go with that.

If you want to stick with partitions, I suggest you try -m MFP+MERGE, or maybe other options in advanced tutorial.

• 0 views • link
written by Mensur Dlakic Mensur Dlakic

Log in to answer this question.

More posts like this