Dear Sergei and other smart people on here,

I was wondering if you could answer a few questions about my analyses. I sent you an email before i discovered this forum today, so please feel free to ignore it! 

1. I have some alignments with gene orthologues and paralogues together. I believe there has been a duplication followed by positive selection just along that one branch, following the duplication. Can I use GABranch to test this?
What about if recombination has been detected using GARD?

2. Are there biological processes other than recombination that could lead to break points in the alignment? I am not entirely sure about recombination in this case.

3. What is the minimum number of sequences to get meaningful results using both GARD and selection detection programs? I have some alignments with only 3 or 4 sequences, about 300 - 800 bp.

Many thanks for your time! 

Many thanks for your reply, its very useful.

Do you know of any methods that attempt to distinguish between the signal/s of recombination and those of local rate variation or heterotachy?

Is this theoretically possible?
Shauna

Hi there,

Regarding question 3), about how to analyse very small alignments, do you mean they can only be analysed by distance methods ie in Mega, or do you mean they can be analsed by likelihood methods but only as implemented in PARRIS, for example?

I am also interested in your opinion about our GAbranch results which clearly show a difference in selective regime on a branch after a gene duplication event, as might be expected.
However, the alignment has some recombination as predited by GARD. I would like to know how to determine whether the recombination is real and whether it has potentially impacted the detection of selection on the particular branch after duplication?

thanks

Hi again,
I think it should be possible to modify the PARRIS script to assess branch-specific rates, using the ReplicateConstraint() command on the respective branches in each tree.  Probably somewhere around line 697 in PARRIS.bf:

Code:

ExecuteCommands ("ReplicateConstraint (\"this1.?.synRate:=this2.?.t__/codonFactor\",givenTree_" + fileID + ",nucTree_" + fileID + ");");
 

and something like what's done in YangNielsenBranchSite2005.bf:

Code:

	/* constrain the foreground branch first */
	for (bc = 0; bc < Columns (stOption); bc = bc + 1)
	{
		ExecuteCommands ("givenTree."+choiceMatrix[stOption[bc]][0]+".nonSynRate:=omega_FG*givenTree."+choiceMatrix[stOption[bc]][0]+".synRate;");
	}

	/* constrain other branches next */
	ReplicateConstraint ("this1.?.nonSynRate:=omega_BG*this2.?.synRate",givenTree,givenTree);
 

I'll get back to this later, it's getting late..

- Art.

Hi Shauna,

I've got a kid myself (11 months old) so I understand completely

This is a tall order.  The PARRIS batch file is bucket-loads of code to absorb and modifying it to address your question will take at least several days to do even for an experienced programmar.

I suppose one way to start is if you could send me a toy data set that I could muck about with in PARRIS under default settings before I break it. 

- Art.

Okay, I've merged the PARRIS.bf and YangNielsenBranchSite2005.bf files.  You need to put the attached file into the /TemplateBatchFiles directory for it to work properly.  The file assumes a fully-reversible nucleotide substitution model (you can edit this on line 22) --- this might be an over-parameterization for your data.  

When you run the file, it will ask you how many files to input.  Enter '1' in the console window.  Select a file containing the NEXUS output from a GARD analysis of your data.  It should contain multiple partitions.  For each partition and tree, you will be asked to define foreground and background branches.  These two levels are differentiated by their dN/dS rate ratio.  Note that you can select MORE THAN ONE branch to place in the foreground.

The analysis depends on your being able to define meaningful foreground and background for every partition and tree!  I don't have any way of automating this  

Then HyPhy will optimize the likelihood function consisting of the tree whose synonymous rates (branch lengths) are constrained to nucleotide trees estimated from each partition and substitution rate biases estimated from the entire alignment.

Post-processing (i.e. interpretation of results) that were present in the NielsenYangBranchSite2005.bf batch file have been stripped out, because they would barf on the multiple partitions.  

Dear Art,

I ran the analysis with the data set, and it worked perfectly!
now I have a question about the interpretation of the results.

Hi there,

I have done LRTs on the results from the different models and the results all make sense. In one dataset I had a FG/BG omega of 11, but when I did the LRT with a null model assuming no positive selection, it could not be rejected. I am guessing this is because the other branches in that tree were too variable in their selective regimes. Still, for the other alignments the results made sense.

I am just finishing the paper and have acknowledged your help. Many thanks