BUSTED-PH results not symmetric when exchanging foreground and background labels

[tlinksvayer]

Dear @spond,

When using BUSTED-PH to test for episodic diversifying positive selection in certain foreground vs background branches, we were surprised to see that the results in foreground and background branches were not "symmetric" when we switched the labels and re-ran the test. For example, a gene might be identified as experiencing significant selection in foreground but not background branches, but when we reverse the definitions of foreground and background, the gene is not identified as experiencing significant selection in background but not foreground. Do I understand correctly that BUSTED-PH explicitly tests for selection in both foreground and background branches (and the difference between foreground and background), and if so, shouldn't we expect symmetry in the results when labels are switched?

thanks!
Tim

[spond]

Dear @tlinksvayer,

It should be symmetric: if FG is found under selection in the direct labeling then it should also be found under selection if it should be found under selection as BG as well.

Do you have examples? If you see strong asymmetry, that's probably due to a convergence issue.

Best,
Sergei

[tlinksvayer]

Dear @spond,

Thanks very much for your quick response and confirming that we can expect symmetry in the FG and BG results when labels are switched.

Sorry for my slow response back: We do see cases of asymmetrical results, where there are qualitative differences when labels are switched between FG and BG. I'll email you some examples.

thanks again!
Tim