Can we estimate branch lengths of a consensus tree using different models?

[wdlingit]

For my first few runs of iqtree2 (v2.3.5), I always see WARNING: Default model LG may be under-fitting. Use option '-m TEST' to determine the best-fit model. in the log when I applied -m LG. I made a run with -m TEST and iqtree2 tested number of models. From the test results, there are two lines:

 No. Model         -LnL         df  AIC          AICc         BIC
  1  LG            77264078.129 187 154528530.258 154528530.285 154530921.295

and

 No. Model         -LnL         df  AIC          AICc         BIC
 72  Q.yeast+F+I+G4 69530880.657 208 139062177.314 139062177.347 139064836.863

It seems that the yeast model is (biologically) reasonable to me and got a better -LnL value so I decided to apply -m Q.yeast+F+I+G4. However, when I applied a command (*) like iqtree3 -redo -T AUTO --threads-max 40 -te phylo3_3add.alltrees.contree -pre phylo3_3add.alltrees.contree -m Q.yeast -s MAFFT_3add/, I got the same warning message WARNING: Default model LG may be under-fitting. Use option '-m TEST' to determine the best-fit model..

The final log-likelihood of iqtree3 -redo -te phylo3_3add.alltrees.contree -pre phylo3_3add.alltrees.contree -m Q.yeast -s MAFFT_3add/ is

1. Initial log-likelihood: -78160094.859
Optimal log-likelihood: -77283987.706

The final log-likelihood with iqtree3 -redo -te phylo3_3add.alltrees.contree -pre phylo3_3add.alltrees.contree -m LG -s MAFFT_3add/ is

1. Initial log-likelihood: -78160094.859
Optimal log-likelihood: -77283987.706

They are the same so it seems that -m is not working in the combination of -te phylo3_3add.alltrees.contree -m Q.yeast. phylo3_3add.alltrees.contree is a rooted tree made by following iqtree FAQ for standard bootstrap, where individual trees are rooted gene trees made by OrthoFinder. I think it would be reasonable to apply an appropriate model for estimating branch lengths of a given consensus tree. Or, it is actually not necessary to switch models if we just want to estimate branch lengths?

*: switched to iqtree3 (downloaded v3.0.1 binary for intel linux) for making sure -te is a working option because I didn't see -te when apply iqtree2 -h.

[evolbeginner]

Hi there, would you pls mind providing your data? a few alignments plus the consensus tree would help. thx.

sishuo

[wdlingit]

Hi @evolbeginner Thank you for the reply. I will have to ask my colleague about sharing data, and maybe randomly pick some of them and reproduce what I saw. Meanwhile, let me post some more complete logs from the two mentioned iqtree3 runs.

IQ-TREE version 3.0.1 for Linux x86 64-bit built May  5 2025
Developed by Bui Quang Minh, Thomas Wong, Nhan Ly-Trong, Huaiyan Ren
Contributed by Lam-Tung Nguyen, Dominik Schrempf, Chris Bielow,
Olga Chernomor, Michael Woodhams, Diep Thi Hoang, Heiko Schmidt

Host:    comp02 (AVX512, FMA3, 125 GB RAM)
Command: ../scripts/iqtree-3.0.1-Linux-intel/bin/iqtree3 -redo -T AUTO --threads-max 40 -te phylo3_3add.alltrees.contree -pre phylo3_3add.alltrees.contree -m Q.yeast -s MAFFT_3add/
Seed:    691911 (Using SPRNG - Scalable Parallel Random Number Generator)
Time:    Wed Jan 14 17:10:02 2026
Kernel:  AVX+FMA - auto-detect threads (40 CPU cores detected)
(...)
NOTE: minimal branch length is reduced to 0.000000037879 for long alignment
Checking for duplicate sequences: done in 2.09861 secs using 122.8% CPU
Reading input tree file phylo3_3add.alltrees.contree ... rooted tree
WARNING: Default model LG may be under-fitting. Use option '-m TEST' to determine the best-fit model.

NOTE: 19898 MB RAM (19 GB) is required!
Measuring multi-threading efficiency up to 40 CPU cores
2 trees examined
Threads: 1 / Time: 342.060 sec / Speedup: 1.000 / Efficiency: 100% / LogL: -133373619
Threads: 2 / Time: 210.961 sec / Speedup: 1.621 / Efficiency: 81% / LogL: -133373619
Threads: 3 / Time: 152.299 sec / Speedup: 2.246 / Efficiency: 75% / LogL: -133373619
Threads: 4 / Time: 118.794 sec / Speedup: 2.879 / Efficiency: 72% / LogL: -133373619
Threads: 5 / Time: 96.461 sec / Speedup: 3.546 / Efficiency: 71% / LogL: -133373619
Threads: 6 / Time: 82.788 sec / Speedup: 4.132 / Efficiency: 69% / LogL: -133373619
Threads: 7 / Time: 74.737 sec / Speedup: 4.577 / Efficiency: 65% / LogL: -133373619
Threads: 8 / Time: 66.497 sec / Speedup: 5.144 / Efficiency: 64% / LogL: -133373619
Threads: 9 / Time: 62.070 sec / Speedup: 5.511 / Efficiency: 61% / LogL: -133373619
Threads: 10 / Time: 62.942 sec / Speedup: 5.435 / Efficiency: 54% / LogL: -133373619
Threads: 11 / Time: 53.815 sec / Speedup: 6.356 / Efficiency: 58% / LogL: -133373619
Threads: 12 / Time: 53.873 sec / Speedup: 6.349 / Efficiency: 53% / LogL: -133373619
Threads: 13 / Time: 54.648 sec / Speedup: 6.259 / Efficiency: 48% / LogL: -133373619
Determining AUTO threadcount: done in 1441.78 secs using 441.3% CPU
BEST NUMBER OF THREADS: 11

Estimate model parameters (epsilon = 0.010)
1. Initial log-likelihood: -78160094.859
Optimal log-likelihood: -77283987.706
Parameters optimization took 1 rounds (100.209 sec)
Wrote distance file to... 
BEST SCORE FOUND : -77283987.706
Total tree length: 7.726

Total number of iterations: 0
CPU time used for tree search: 0.000 sec (0h:0m:0s)
Wall-clock time used for tree search: 0.000 sec (0h:0m:0s)
Total CPU time used: 7444.267 sec (2h:4m:4s)
Total wall-clock time used: 1550.722 sec (0h:25m:50s)

Analysis results written to: 
  IQ-TREE report:                phylo3_3add.alltrees.contree.iqtree
  Maximum-likelihood tree:       phylo3_3add.alltrees.contree.treefile
  Screen log file:               phylo3_3add.alltrees.contree.log

Date and Time: Wed Jan 14 17:36:21 2026

IQ-TREE version 3.0.1 for Linux x86 64-bit built May  5 2025
Developed by Bui Quang Minh, Thomas Wong, Nhan Ly-Trong, Huaiyan Ren
Contributed by Lam-Tung Nguyen, Dominik Schrempf, Chris Bielow,
Olga Chernomor, Michael Woodhams, Diep Thi Hoang, Heiko Schmidt

Host:    comp11 (AVX512, FMA3, 125 GB RAM)
Command: ../scripts/iqtree-3.0.1-Linux-intel/bin/iqtree3 -redo -T AUTO --threads-max 40 -te phylo3_3add.alltrees.contree -pre phylo3_3add.alltrees.contree -m LG -s MAFFT_3add/
Seed:    219755 (Using SPRNG - Scalable Parallel Random Number Generator)
Time:    Wed Jan 14 18:01:43 2026
Kernel:  AVX+FMA - auto-detect threads (40 CPU cores detected)
(...)
NOTE: minimal branch length is reduced to 0.000000037879 for long alignment
Checking for duplicate sequences: done in 2.07343 secs using 121.5% CPU
Reading input tree file phylo3_3add.alltrees.contree ... rooted tree
WARNING: Default model LG may be under-fitting. Use option '-m TEST' to determine the best-fit model.

NOTE: 19898 MB RAM (19 GB) is required!
Measuring multi-threading efficiency up to 40 CPU cores
2 trees examined
Threads: 1 / Time: 362.434 sec / Speedup: 1.000 / Efficiency: 100% / LogL: -135220688
Threads: 2 / Time: 206.375 sec / Speedup: 1.756 / Efficiency: 88% / LogL: -135220688
Threads: 3 / Time: 140.974 sec / Speedup: 2.571 / Efficiency: 86% / LogL: -135220688
Threads: 4 / Time: 111.393 sec / Speedup: 3.254 / Efficiency: 81% / LogL: -135220688
Threads: 5 / Time: 90.034 sec / Speedup: 4.026 / Efficiency: 81% / LogL: -135220688
Threads: 6 / Time: 82.328 sec / Speedup: 4.402 / Efficiency: 73% / LogL: -135220688
Threads: 7 / Time: 70.566 sec / Speedup: 5.136 / Efficiency: 73% / LogL: -135220688
Threads: 8 / Time: 60.436 sec / Speedup: 5.997 / Efficiency: 75% / LogL: -135220688
Threads: 9 / Time: 57.453 sec / Speedup: 6.308 / Efficiency: 70% / LogL: -135220688
Threads: 10 / Time: 55.561 sec / Speedup: 6.523 / Efficiency: 65% / LogL: -135220688
Threads: 11 / Time: 49.998 sec / Speedup: 7.249 / Efficiency: 66% / LogL: -135220688
Threads: 12 / Time: 43.447 sec / Speedup: 8.342 / Efficiency: 70% / LogL: -135220688
Threads: 13 / Time: 41.338 sec / Speedup: 8.768 / Efficiency: 67% / LogL: -135220688
Threads: 14 / Time: 40.033 sec / Speedup: 9.053 / Efficiency: 65% / LogL: -135220688
Threads: 15 / Time: 39.567 sec / Speedup: 9.160 / Efficiency: 61% / LogL: -135220688
Threads: 16 / Time: 36.865 sec / Speedup: 9.831 / Efficiency: 61% / LogL: -135220688
Threads: 17 / Time: 38.781 sec / Speedup: 9.346 / Efficiency: 55% / LogL: -135220688
Threads: 18 / Time: 35.933 sec / Speedup: 10.086 / Efficiency: 56% / LogL: -135220688
Threads: 19 / Time: 35.986 sec / Speedup: 10.072 / Efficiency: 53% / LogL: -135220688
Threads: 20 / Time: 33.603 sec / Speedup: 10.786 / Efficiency: 54% / LogL: -135220688
Threads: 21 / Time: 33.342 sec / Speedup: 10.870 / Efficiency: 52% / LogL: -135220688
Threads: 22 / Time: 33.226 sec / Speedup: 10.908 / Efficiency: 50% / LogL: -135220688
Determining AUTO threadcount: done in 1716.42 secs using 651.4% CPU
BEST NUMBER OF THREADS: 20

Estimate model parameters (epsilon = 0.010)
1. Initial log-likelihood: -78160094.859
Optimal log-likelihood: -77283987.706
Parameters optimization took 1 rounds (66.718 sec)
Wrote distance file to... 
BEST SCORE FOUND : -77283987.706
Total tree length: 7.726

Total number of iterations: 0
CPU time used for tree search: 0.000 sec (0h:0m:0s)
Wall-clock time used for tree search: 0.000 sec (0h:0m:0s)
Total CPU time used: 12497.631 sec (3h:28m:17s)
Total wall-clock time used: 1791.282 sec (0h:29m:51s)

Analysis results written to: 
  IQ-TREE report:                phylo3_3add.alltrees.contree.iqtree
  Maximum-likelihood tree:       phylo3_3add.alltrees.contree.treefile
  Screen log file:               phylo3_3add.alltrees.contree.log

Date and Time: Wed Jan 14 18:32:04 2026

[evolbeginner]

thank you. seems to me a bit bizarre that -m LG doesn't work on a fixed topo tree. perhaps you can try simulating some alignments along the tree using your input fixed tree by alisim and see what happens? not sure.

sishuo