Skip to content

Frequently Asked Questions

minh edited this page Jul 15, 2016 · 51 revisions

For common questions and answers.

Table of Contents

How do I interpret ultrafast bootstrap (UFBoot) support values?

The ultrafast bootstrap (UFBoot) feature (-bb option) was published in (Minh et al., 2013). One of the main conclusions is, that UFBoot support values are more unbiased: 95% support correspond roughly to a probability of 95% that a clade is true. So this has a different meaning than the normal bootstrap supports (where you start to believe in the clade if it has >80% BS support). For UFBoot, you should only start to believe in a clade if its support is >= 95%. Thus, the interpretations are different and you should not compare BS% with UFBoot% directly.

Moreover, it is recommended to also perform the SH-aLRT test (Guindon et al., 2010) by adding -alrt 1000 into the IQ-TREE command line. Each branch will then be assigned with SH-aLRT and UFBoot supports. One would typically start to rely on the clade if its SH-aLRT >= 80% and UFboot >= 95%.

How does IQ-TREE treat gap/missing/ambiguous characters?

Gaps (-) and missing characters (? or N for DNA alignments) are treated in the same way as unknown characters, which represent no information. The same treatment holds for many other ML software (e.g., RAxML, PhyML). More explicitly, for a site (column) of an alignment containing AC-AG-A (i.e. A for sequence 1, C for sequence 2, - for sequence 3, and so on), the site-likelihood of a tree T is equal to the site-likelihood of the subtree of T restricted to those sequences containing non-gap characters (ACAGA).

Ambiguous characters that represent more than one character are also supported: each represented character will have equal likelihood. For DNA the following ambigous nucleotides are supported according to IUPAC nomenclature:

Nucleotide Meaning
R A or G (purine)
Y C or T (pyrimidine)
W A or T (weak)
S G or C (strong)
M A or C (amino)
K G or T (keto)
B C, G or T (next letter after A)
H A, C or T (next letter after G)
D A, G or T (next letter after C)
V A, G or C (next letter after T)
?, -, ., ~, O, N, X A, G, C or T (unknown; all 4 nucleotides are equally likely)

For protein the following ambiguous amino-acids are supported:

Amino-acid Meaning
B N or D
Z Q or E
J I or L
U unknown AA (although it is the 21st AA)
?, -, ., ~, * or X unknown AA (all 20 AAs are equally likely)

Can I mix DNA and protein data in a partitioned analysis?

Yes! You can specify this via a NEXUS partition file. In fact, you can mix any data types supported in IQ-TREE, including also codon, binary and morphological data. To do so, each data type should be stored in a separate alignment file (see also Partitioned analysis with mixed data). As an example, assuming dna.phy is a DNA alignment and and prot.phy is a protein alignment. Then a partition file mixing two types of data can be specified as follows:

#nexus
begin sets;
    charset part1 = dna.phy: 1-100 201-300;
    charset part2 = dna.phy: 101-200;
    charset part3 = prot.phy: 1-150;
    charset part4 = prot.phy: 151-400;
    charpartition mine = HKY:part1, GTR+G:part2, WAG+I+G:part3, LG+G:part4;
end;

NOTE: The site count for each alignment should start from 1, and not continue from the last position of a previous alignment (e.g., see part3 and part4 declared above).

What is the interpretation of branch lengths when mixing codon and DNA data?

When mixing codon and DNA data in a partitioned analysis, the branch lengths are interpreted as the number of nucleotide substitutions per nucleotide site! This is different from having only codon data, where branch lengths are the number of nucleotide substitutions per codon site (thus typically 3 times longer than under DNA models).

Note that if you mix codon, DNA and protein data, the branch lengths are then the number of character substitutions per site, where character is either nucleotide or amino-acid.

What is the purpose of the composition test?

At the beginning of each run, IQ-TREE performs a composition chi-square test for every sequence in the alignment. The purpose is to test for homogeneity of character composition (e.g., nucleotide for DNA, amino-acid for protein sequences). A sequence is denoted failed if its character composition significantly deviates from the average composition of the alignment.

More specifically, for each sequence, compute:

chi2 = \sum_{i=1}^k (O_i - E_i)^2 / E_i

where k is the size of the alphabet (e.g. 4 for DNA, 20 for amino acids) and the values 1 to k correspond uniquely to one of the characters. O_i is the character frequency in the sequence tested. E_i is the overall character frequency from the entire alignment.

Whether the character composition deviates significantly from the overall composition is done by testing the chi2 value using the chi2-distribution with k-1 degrees of freedom (df=3 for DNA or df=19 for amino acids). By and large it is a normal Chi^2 test.

This test should be regarded as an explorative tool which might help to nail down problems in a dataset. One would typically not remove failing sequences by default. But if the tree shows unexpected topology the test might point in direction of the origin of the problem.

Furthermore, please keep in mind, this test is performed at the very beginning, where IQ-TREE does not know anything about partitions yet. That means, it might be more reasonable to test this separately for each partition in a partition analysis. Here, one might want to be able to decide whether some partitions should better be discarded if it is hard to find a composition representing the sequences in the partition. Or on the other hand if a sequence fails for many partitions and show very unexpected phylogenetic topologies, try without it.

In any case if one has sequences that fail, it is recommended to always check the alignment (something one should always do anyway), especially if they have been collected and produced automatically.

What is the good number of CPU cores to use?

IQ-TREE can utilize multicore machines to speed up the analysis via -nt option. However, it does not mean that using more cores will always result in less running time: if your alignment is short, using too many cores may even slow down the analysis. This is because IQ-TREE parallelizes the likelihood computation along the alignment. Thus, the parallel efficiency is only increased with longer alignments. As a rule of thumb:

  • For DNA data: 1 core for every 500 alignment sites. For example, if your alignment length is 1300 bp, use at most 3 cores.
  • For protein data: 1 core for every 100 sites.
  • For codon data: 1 core for every 50 sites.
  • For binary data: 1 core for every 1000 sites.

Please note that the more proper way is to actually run some quick analysis with increasing number of cores for your data and choose the one that best balances the trade-off between number of cores and waiting time. See this tutorial for more details.

How do I speed up the standard bootstrap?

The standard bootstrap is rather slow and may take weeks/months for large data sets. One way to speed up is to use the multicore version. However, this only works well for long alignments (see What is the good number of CPU cores to use?). Another way is to use many machines or a computing cluster and split the computation among the machines. To illustrate, you want to perform 100 bootstrap replicates and have 5 PCs, each has 4 CPU cores. Then you can:

  1. Perform 5 independent bootstrap runs (each with 20 replicates) on the 5 machines with 5 prefix outputs (such that output files are not overwritten). For example:

     iqtree-omp -nt 4 -s input_alignment -bo 20 ... -pre boot1
     iqtree-omp -nt 4 -s input_alignment -bo 20 ... -pre boot2
     iqtree-omp -nt 4 -s input_alignment -bo 20 ... -pre boot3
     iqtree-omp -nt 4 -s input_alignment -bo 20 ... -pre boot4
     iqtree-omp -nt 4 -s input_alignment -bo 20 ... -pre boot5
    
  2. Once all 5 runs finished, combine the 5 .boottrees file into one file (e.g. by cat command under Linux):

     cat boot*.boottrees > alltrees
    
  3. Construct a consensus tree from the combined bootstrap trees:

     iqtree -con -t alltrees
    

    The consensus tree is then written to .contree file.

  4. You can also perform the analysis on the original alignment:

     iqtree-omp -nt 4 -s input_alignment ...
    

    and map the support values onto the obtained ML tree:

     iqtree -sup input_alignment.treefile -t alltrees 
    

    The ML tree with assigned bootstrap supports is written to .suptree file.

Clone this wiki locally