-
Notifications
You must be signed in to change notification settings - Fork 1
Frequently Asked Questions
For common questions and answers.
Table of Contents
- How do I interpret ultrafast bootstrap (UFBoot) support values?
- How does IQ-TREE treat gap/missing/ambiguous characters?
- Can I mix DNA and protein data in a partitioned analysis?
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%.
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) |
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 partition file format).
NOTE: The site count for each alignment should start from 1, and not continue from the last position of a previous alignment.
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.
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.
Typically one would call this a Chi^2 goodness-of-fit test computing:
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 character frequency expected from the ‘master’ distribution (e.g. the overall frequencies - depends on what one is using).
Whether the character composition deviates significantly for the ‘master’ distribution 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.
Copyright (c) 2010-2022 IQ-TREE development team.
- First example
- Model selection
- New model selection
- Codon models
- Binary, Morphological, SNPs
- Ultrafast bootstrap
- Nonparametric bootstrap
- Single branch tests
- Partitioned analysis
- Partitioning with mixed data
- Partition scheme selection
- Bootstrapping partition model
- Utilizing multi-core CPUs
- Tree topology tests
- User-defined models
- Consensus construction and bootstrap value assignment
- Computing Robinson-Foulds distance
- Generating random trees
- Estimating amino acid substitution models
- DNA models
- Protein models
- 3Di and TEA models
- Codon models
- Binary, morphological models
- Ascertainment bias correction
- Rate heterogeneity
- Counts files
- First running example
- Substitution models
- Virtual population size
- Sampling method
- Bootstrap branch support
- Interpretation of branch lengths