Interpreting Hyphy Testbranchdnds.Bf

Question

[Rvosa]

Hi,

I'm trying to accomplish the following: find out the dN/dS ratio on each branch of a tree, and whether some ratios are significantly different from others, using HyPhy.

Here are the steps I'm thinking about taking (in the HyPhy GUI):

• Standard Analyses > Positive Selection > TestBranchDNDS.bf
• Genetic code is Universal (vertebrate nuclear)
• load a nexus file with an alignment and a tree
• No site-by-site rate variation
• Default nucleotide substitution model (MG94xHKY85, only corrects for Ti/Tv)
• Default amino acid class model (all changes are equivalent)
• Use the provided tree from the file
• Select all branches from the list
• Parameters to test: 1 (non-synonymous rate parameter nsClass1)

My understanding is that this will run a likelihood ratio test where, under the null hypothesis, the non-synonymous rate parameter is constrained to be equal over all branches, and the alternative hypothesis allows it to vary freely.

Consequently, the number of parameters to estimate in the alternative hypothesis equals the number of branches on the tree, and the difference in number of parameters (for the LRT) is number of branches - 1.

OK, so when I look at the results, the log likelihood for null is -3566.06313872664, and for the alternative it's -3545.22024602892. This makes sense because the alternative hypothesis allows a better fit to the data. In this particular case, p << 0.01, so I conclude that at least some of the branches require a different dN rate class than some others (correct?).

But there are some things that I don't understand:

1. is this a reasonable thing to do? I understand that the nucleotide and amino acid models could be more specified, but beyond that: can I constrain all branches in the null and none in the alternative and will that give me useful results?
2. what's shared parameter "AC"?
3. what are the branch lengths on the null and alternative trees that are logged?
4. what are the branch parameter estimates listed as givenTree.${NODELABEL}.nsClass1?
5. basically, can I now get the dN/dS ratio for each branch out of what was produced?

Thanks. I'm a HyPhy newbie so this is all a bit hard to make sense out of.

[Dave Lunt]

A v quick post, not really answering your questions, but I've found the HYPHY discussion boards really useful for these sorts of questions it would definitely be worth cross posting and trying to synthesise answers back here maybe?

http://www.hyphy.org/cgi-bin/hyphy_forums/yabb.pl