I would like to check my interpretation of GARD analysis on a set of toxin sequences. My whole file is too large to run on the webserver, and attempts to install HyPhy locally have failed (even on a Linux box that has an MPI library installed I get the message"requires an MPI environment to run"). I have therefore split my sequences up into subfiles. Reassuringly, the analyses have tended to come up with the same number of significant breakpoints, many of which are in the same locations. I have also analysed the whole file using RDP3 and the breakpoints identified in these tend to confirm the identified breakpoints. I then checked and removed all the identified recombinants and ran the GARD analysis again but was surprised to see that the same results were obtained. Checking the position of the identified breakpoints, I have noticed that they correspond (very precisely in some cases) to intron/exon boundaries. Essentially, the breakpoints have divided my alignment into the following regions: 5'UTR and translated signal region (there is also a breakpoint within this region but it is the least stable in position between different analyses and also not very strongly supported), the coding region of exon2 and a bit of intron2, first half of intron2 (very long), second half of intron 2, exon3 (coding), intron3, exon4 (coding and 3'UTR). My interpretation is therefore that this is not the result of true recombination (with the possible exception of the breakpoint in the middle of intron2) but instead reflects positive selection on the protein coding region. I plan therefore to separate introns and exons and analyse them separately, using the introns and 5' and 3'UTR+ signal region for estimating the phylogenetic history of the alleles (having first removed the clearly identified recombinants) and using the protein coding region for studying selection. Does this sound like a justifiable approach? Do you think it would be worth repeating the analysis with the protein coding regions excised?