My input alignment was screened for recombination using GARD (Genetic Algorithm for Recombination and Detection) and 2 recombination breakpoints were detected which resulted in 3 trees.

For subsequent positive selection analyses using codeml (PAML), I used the GARD generated nexus tree and the multiple alignment file. PAML uses all 3 provided trees and runs the m2a/m1a and m8/m7 models and identifies positively selected sites for each given tree.

You get positively selected sites (from BEB method) independently for m2a/m1a and m8/m7 for every tree.

example:

m2a/m1a - tree1
m2a/m1a - tree2
m2a/m1a - tree3

m8/7 for all the above trees separately.

In total 6 combinations as above for arriving at positively selected sites. many of these sites overlap and however there are some sites which are unique to a given model and a given tree.

Now, the question is, should we take consensus between m2a/m1a and m8/m7 and all trees? i.e., sites which are reported to be positively selected by all these 6 combinations? to be conservative? what people practically do?