Likelihood ratio tests were performed to assess whether permitting some codons to evolve under positive selection gives a significantly better fit to the data than models where positive selection is not allowed [ 53, 54].
Likelihood ratio tests revealed that a model that accommodates positive selection is a significantly better fit to the data than a model that does not allow for positive selection for VP35-like sequences.
Among the genes for which the diversifying selection models showed a significantly better fit of the data, the proportion of sites under diversifying selection ranged from 0.1% to 14.6% (model M2a) and 0.1% to 29.5% (model M8).
An LRT showed that the M8 model allowing positive selection was a significantly better fit to the data than M7: 2Δl = 2 -3260.66 -(-3253.50) = 16.32, P < 0.001 with 2 degrees of freedom (df).
When the CM-based ranking is used to rank the RF templates in cases where the highest predicted TM-distance is >0.5 (19 cases in the RA benchmark) the model accuracy improves in 10 cases, leading to the selection of a significantly better model in 5 cases (PDB ID: 1BQL, 1IQD, 1FBI and 1HZH, 4.3, 3.8, 7.7, 5.0 and 3.7 Å RMSD to 2.1, 2.1, 3.7 and 2.4 Å RMSD, respectively).
Twenty-five genes, out of the 372 tested, showed a significantly better fit to the models under positive selection (LRT; P < 0.05).
In all cases the positive selection model was a significantly better fit (p < 0.05), and individual codons assigned to the dN/dS > 1 class with high posterior probabilities (P > 0.85 by Bayes Emperical Bayes [ 56]) were analyzed.
For the gene family tested, the alternative model (specifically testing for positive selection) was not a significantly better fit (2λ < 2.71, df = 1, p > 0.05) than the null model (Additional file 3).
The paralog models (Mp1, Mp2, and Mp3; see Methods) that allow for paralog-specific differences in selection pressure provided a significantly better explanation of the data (p < 0.0001, [see additional file 5: thane_S4]) than did the one-ratio model (M0) assuming no differences in selection pressure among Epd paralog groups.
The positive selection models were again a significantly better fit to the sequence set than the null models, with a p value ≤ 0.0003 (Table 1).
