To refer to this article use this url: http://ctoz.nl/vol86/nr02/a01
Test of model fit
Results of the model selection based on Bayes factors (BF, Table 1) indicated that the best fitting model employed a combination of a relaxed exponential clock and a Birth-Death model of tree diversification. We used this combination of priors for subsequent analyses (Tables 2, 3). The difference in BF to other model combinations was quite low (BF 2.4 over the second ranking model), but enough to set this model as favoured. In general, model combinations involving a relaxed exponential were “strongly” favoured over the relaxed lognormal, the latter being in turn “very strongly” favoured over the strict and the random clock models. As for the tree priors, the difference between the Coalescent and the three Birth-Death type processes we tested was very low.
Because the difference of the marginal likelihood using different priors was extremely low, we advocated great care in interpreting the result of our model selection: Table 1 is however a good indication for future tardigrade clock studies using RNA sequences, rather than as decision maker to discriminate among competing time estimates.
Table 2. Molecular dating using two different sets of calibration priors. Analyses were conducted using the complete dataset and the most fitting relaxed clock and tree prior as defined in Table 1 (relaxed exponential clock plus Birth Death process). For each of the calibration sets, we provide the mean estimates in millions of years for four nodes of interest; the heights of the 95% Highest Posterior Density are in parenthesis. |
Table 3. Molecular clock analyses using reduced-datasets. Analyses were conducted using the most fitting relaxed clock and tree prior as defined in Table 1 (relaxed exponential clock plus Birth Death process). For each of the calibration-sets, the mean estimates in millions of years for four nodes of interest are provided; the heights of the 95% Highest Posterior Density are in parenthesis. |