Dear Sergei,

Given that we (and others, by the looks of it) are looking for compartmentalization (spatial structure to you non-HIV folks) of sequences, how about we add in a module into the standard analyses to do compartmentalization analysis?

Possibilities include:
1. Calculation of Fst, and testing using AMOVA, etc. based on a distance matrix. One could obtain this from scratch from the sequences, by loading a distance matrix, or calculating distances from a tree.
2. The Slatkin-Maddison test.
3. Bootstrap test of clustering by population.
4. Testing for different dN/dS within and between populations (assuming 100% clustering).

I'm sure there are more if we think about it a little. What do you think?

Best,
Simon

Dear Sergei,

What with the HIV Dynamics meeting coming up, it's probably a good idea to package up the compartment stuff (see above). In addition, there are a few PositiveSelection analyses that it would be nice to tweak to allow more complex hypotheses regarding the role of selection in compartmentalization to be conducted a little more easily. For example:

1. The ability to select multiple classes of branch in TestBranchDnDs.

2. Given a sample of sequences from (say) two compartments, reconstruct the migration history using parsimony and spool a list of branches which comprise of (a) within compartment branches and (b) between compartment branches.

3. The ability to include partitions of branches in a HyPhy nexus file, in the same way as partitions of sites are now done.

4. In the output of TestBranchDNDS, to report the branches tested in the analysis on the console.

What do you think?

Best
Simon

Dear Simon,

Quote:

This feature has been added to StandardAnalyses->Compartmentalization->BranchClassDNDS.bf

Quote:

Such a list is produced by the Slatkin-Maddison test and can be used as input to StandardAnalyses->Compartmentalization->BranchClassDNDS.bf

Cheers,
Sergei

Dear Rose,

It all depends on the format of your sequence names. Let's say the sequences look like this:

1C
2C
3C
1P
2P
3P

with the last letter indicating which population the sequence has come from. The regexp P$ (last letter is P) will place all the sequences for which the regexp is true into the first set.

If your sequences begin with a certain letter:

C1
C2
C3
P1
P2
P3

then the regexp ^P (first letter is P) will do the trick.

Regular expressions are extremely useful things; there are plenty of regex debuggers out there. Have a look at Kodos, for example (http://kodos.sourceforge.net/).

Best,
Simon

Dear Rose,

If you're interested in whether selection differs on a given branch, or set of branches, relative to the rest, you can either do a 'Branch-Site test', which is not currently available in HyPhy (it can easily be added), in which a distribution of rates across sites is assumed, or you can adopt a fixed effects approach, where the rates at each site are fitted separately. The latter approach (which, for reasons I won't go into here, may be preferable to the first approach) can be done as part of the 'Positive Selection' analyses.

So, if you want to know whether selection at a site along a branch which separates two compartments (or all those that separate the compartments), do the following:

Standard Analyses>
PositiveSelection>
QuickSelectionDetection>
(Choose your genetic code e.g. universal)
(New analysis)
(Specific codon file)
(Choose model options (e.g. the default is MG94 X HKY85))
(Select tree file)
(Save nucleotide fit to: (e.g mynucfit))
(dN/dS bias parameter options - e.g. estimate + C.I.)
(Ancestor counting options - two rate FEL (THIS IS THE IMPORTANT BIT))

You'll then wait a bit, until it asks you for the P-value cutoff for significance (e.g. 0.1). You will then have a choice of branch options. If you choose 'Custom subset', a tree will come up, with the nodes labeled. You can then select one or more branches (in your case, the branches separating the two compartments). This will test whether dN/dS at each site in the alignment is significantly different between the branches you selected and those you have not. If you wish to allow dN/dS to be different in each compartment (i.e., three dN/dS ratios for each site), this will require a custom analysis. Please contact us for help if you need to do this.

Note that the p value that you get from this approach will only tell you whether dN/dS is significantly different between compartments compared to within compartments (assuming that dN/dS is the same within each compartment). If you want to say whether positive selection is occurring at a site, you'll have to get confidence intervals on the estimate for dN/dS at that site.

Best wishes
Simon