huge difference in mean rate between GTR vs HKY substitution model

[Emanuele Berrilli]

Dears,

I am running sum fossil calibrated run in BEAST 1.10 tring to estimate mean rate of some genetic markers. I am using the same dataset with the same fossil calibration and than tested with different subsitution models (HKY and GTR), clock model (I tested both strict clock and uncorrelated relaxed clock lognormal) and partition (123, 12+3 and 1+2+3). I also used marginal likelihood to calculate Bayesian factor to see which is the favorited model. All runs converge quite well, and the model that is preferred according to the BF is the non-partitioned dataset (123) with GTR and uncorrelated clock. However, the doubt that arises in me is that, the mean rate calculated in the runs in which I have not partitioned the codons and when I apply the GTR as an evolutionary model, is a factor of 10 higher than if I apply the HKY: HKY_strict mean rate= 0.0034; GTR_strict mean rate = 0.0371 and HKY_log mean rate = 0.0043; GTR_log mean rate = 0.0168 (in this case a factor 4, but still higher). The pattern of higer values of mean rate with GTR model is repited also with the two other partition scheme (12+3 and 1+2+3) but the different with the HKY model is very slight. Do you have any ideas as to why there is such a big difference in mean rate between GTR and HKY in the non-partitioned scheme? Could that be related to the saturation of the markers I used? Can I take this as a biologically sesate result, or is it an arctifact of the analyses? Thanks in advance for your assistance!

Emanuele