PAML kappa and alpha

Question

[kissaj]

Hello,

What do the "kappa" and "alpha" parameters do in PAML?

Thanks, Andor

Answer 1

[natasha.sernova]

This file is helpful a little bit.

http://abacus.gene.ucl.ac.uk/ziheng/data/pamlDEMO.pdf

kappa=transition/transversion ratio

alpha:see this fragment of PAML-manual:

http://abacus.gene.ucl.ac.uk/software/pamlDOC.pdf

p.36

fix_alpha, alpha: "For codon models, the pair fix_alpha and alpha specify the m odel of gamma rates for sites, in which the relative rate for a site varies among codons according to the gamma distribution, but the ratio stays the same over all sites. This is a lazy extension of the gamma model of rates for nucleotides and amino acids. I don’t like this model and suggest that you use the NSsites models instead (which is specified using the NSsites variable, with fix_alpha = 1, alpha = 0). The program should abort if you specify both NSsites and alpha".

It was discussed in FAQs, see the link and text below:

http://abacus.gene.ucl.ac.uk/software/pamlFAQs.pdf

"How can I estimate an amino acid subs titution matrix from my own data, like mtmam.dat and wag.dat?

You choose the option variable model = 9 :REVaa(nr=189) You can also use fix_alpha = 0 to fit a REVaa+Gamma model. If the data set is large, you might want to prepare an in.codeml file to start the iteration from a good starting point. Here is a way of preparing the file to run the REVaa+Gamma model. No clock (clock = 0) is assumed.

Step 1: Use model = 3 to generate good starti ng values for branch lengths and for alpha.

model = 3 aaRatefile = mtmam.dat for mt proteins or wag.dat for nuclear proteins fix_alpha = 0 alpha = 0.5 method = 1

Step 2: Run REVaa to get good initial values for the substitution rate matrix. Copy the tree with branch lengths obtained in step 1 into the tree file and choose fix_blength =1 to use them as initial values. Other options are

model = 9 fix_alpha = 1 alpha = 1 method = 0

The program will re-estimate the branch lengths as well as the substitution rate parameters in the rate matrix. All those estimates should then be copied into the in.codeml file, plus the alpha estimate from step 1. See notes in step 3.

Step 3: Run REVaa+Gamma. This model involves one more parameter, alpha, than REVaa without the gamma. Copy the parameter estimates under REV aa from step 2 into a file named in.codeml. Add another value for alpha at the end. You had better use the alpha estimate from step 1. Other options are

model = 9 fix_alpha = 0 alpha = 0.5 method = 0

This sounds too complicated. The output from this run include a lower-diagonal matrix, which you can copy into a file suitably named, say, MyFavariateProteinMatrix.dat. Look at mtmam.dat or jones.dat etc. for the format. You can then use model = 2 or 3 for other analysis without re- estimating the matrix all over again. As a check, you should obtain identical (except for small rounding errors) branch lengths and log likelihood values when you use model = 2 or 3 but using the rate matrix you estimated under model = 9".