Dear Sergei,

I am interested in finding sites under positive selection in 7 aligned autologous nef sequences derived from patients before and after therapeutic vaccination with HIV-nef.
Analysis with REL and the recommended codon substitution model HKY85 yielded 10 positively selected and 1 negatively selected site in the post vax sequences with a significance level of 50. The prevax alignment with also 7 seqences resulted in no positively selected site but 9 sites under negative selection. I have taken taken that as a pretty strong indication of positive selection linked to the vaccination.

However, I reanalyzed the sequences using GARD. Now the post-vax alignment resulted in evidence for 3 breakpoints, whereas the pre-vax sequences yielded 2 breakpoints. I analyzed the GARD output files directly in Datamonkey using analogue settings (REL; HKY85) to the first evaluation (without testing for recombination) . Post-vax  (4 partitions) yielded no positively selected sites and one negative selection. Pre vax alignments (3 partitions) resulted in positive selection at 2 codons and 12 instances of negative selection.
I am not sure how to deal with this discrepancy. Are 7 aligned sequences too few for such an analysis? How could I further strengthen our data?

Thanks for your help and suggestions!

Dieter

Dear Sergei,

Thanks for the information!

A brieg summary of the tests you suggested:
PARRIS has not yielded any indicaton of selection.

The c-AIC improvement  in GARD (column were 9,1 with one recomb. site, and 7,6 for 2 and 3 reomb. sites, respectively.

The dN/dS distribution hasn't changed much between analysis with or without GARD.
With only 1 partition (no GARD screening) there are 2 out of the 9 sites with dN/dS >1. With previous GARD screening only one rate combination yields dN/dS > 1.

Here there's no indication for positive selection

Seems like the results are not clear-cut and stable because of the low number of sequences...

Thanks again,

Dieter