Dear Hyphy/Datamonkey

I`m working with PAML, PARRIS and GARD and have some questions about my results. I started my analysis with PAML and found sites among my sequences that I`m working with to be under positive selection (4 under M2, 2 sites under M8). I checked for the interference of recombination using PARRIS and there still seems to be evidence for positive selection (LRT is still significant). Then I screened for recombination sites using GARD and detected one site. I subdivided the alignment in two parts, to analyse both fragments again with PAML and found for one fragment a higher nr of sites under positive selection (6 sites vs 2 earlier found in this region, LRT significant, similar for M2 and M8), for the other fragment I found only 1 (lost 1 site and LRT significant for M1/M2 but no longer under M7/M8). The extra sites found in the first segment were detected in the complete PAML analysis as well, but had then a low probability in the NEB.

My question is which sites I can now believe to be under positive selection? I can understand that if the GARD program detects a recombination site, the parameters as determined in PAML for the complete sequences are not applicable for all sites and that the results on the two seperate fragments are more correct. But then I like to know how one can see how `reliable` the GARD detection of the recombination site is (delta c-AIC between 0 and 1 Bp is 200, so I think very reliable, but what if this difference is 20?). Also, I do not completely understand the table in the results file of PARRIS. What information do the AIC-values for M1 and M2 provide and on what is the given p-value based on? I determined the LRT based on the difference in lnL given for M1 and M2. Sorry, I`m not a statistician, nor bio-informaticus, so I have some difficulties with fully understanding the meaning of a certain lnL- or c-AIC-value.

My question is also more general, because how should I interpretate results from others who have not performed a check for recombination, but did found pos sites under PAML? I can understand that under different models, different conditions are set, so one should present their results in the frame of the model. But also if I find different sites under M2 and M8 of PAML, which sites can I believe to be `really`under diversifying selection?

thanks a lot,
Maria