Re: IQ-TREE “-m C60+LG+G+F” vs “-m LG+C60+G+F”

[Bui Quang Minh]

Hi Ding, 

This is a very good remark, thus I’m replying to the IQ-TREE forum here, in case other users observed this.

IQ-TREE uses alias so that users can specify very complex models with a short model string. In this case, C60 is alias for POISSON+G+FMIX{…}, which specifies the profile mixture model with Gamma rate and importantly, equal amino-acid substitution rate (reflected by POISSON matrix). Therefore:

1. C60+LG+G+F is interpreted as POISSON+G+FMIX{…}+LG+G+F. While parsing this model string, the matrix, which comes first, “wins". Therefore, IQ-TREE applies POISSON matrix.

2. LG+C60+G+F is interpreted as LG+POISSON+G+FMIX{…}+G+F. Therefore the LG matrix is used.

That’s why you should get a much higher log-likelihood with the 2nd model  LG+C60+G+F.

Hope that helps,

Minh

On Nov 20, 2017, at 10:15 AM, Ding He <din...@ebc.uu.se> wrote:

Hi Minh,

I happened to notice last week that the model settings for "C60+LG+G+F” and “LG+C60+G+F” are actually different, as I got drastic different lnL values. I know that the explanation in the manual seems always to be “LG+C60+G+F”, but I think that the model setting could be easily mistakenly wrongly used because of the order difference. 

Best,

/Ding

--

Bui Quang Minh

Center for Integrative Bioinformatics Vienna (CIBIV)

Campus Vienna Biocenter 5, VBC5, Ebene 1

A-1030 Vienna, Austria

Phone: ++43 1 4277 74326

Email: minh.bui (AT) univie.ac.at