We tested the transferability of the models determined with the irrigated field data of our experiment, limiting our evaluation to models that were season independent and explained more than half of the variance of the cluster mean.
Phylogenetic inference was performed using the Bayesian methodology implemented in BEAST v1.7.4 (Drummond et al. 2012), applying the appropriate models determined with jModelTest, or the next most parameterized model available in BEAST.
About a half of the models determined with the method by Nagano et al. (2012) do not have an environmental term and those that do, include the effect of a single climatic factor per gene, with a term that can be non-linear with diurnal changes in sensitivity.
The best model (model 4) as determined by maximal hydrophobic packing in md simulation, model quality (determined with the knowledge-based potential of YASARA2), and energy minimization was manually added onto the model of the HMG box bound to DNA, forming the peptide bond, followed by energy minimizations using the AMBER03 force field in water.
The decomposition process was also followed by Py-GC MS, the activation energy was calculated with isoconversional methods and the reaction model was determined with model-fitting method.
The analog model has been verified on the digital model determined by experiment with using the parametric identification method.
Maximum likelihood analyses were run using RAxML (version 7.0.4, [ 88]) with a JTT+G matrix (model determined by AIC criteria with ProtTest v1.4, [ 89]).
Maximum likelihood analyses were run using RAxML (version 7.0.4) [49] with the optimized protein models (model determined by AIC criteria with ProtTest, [50]).
Trees were constructed using PhyML v3.0 [ 145] with a LG + I + G matrix (model determined by AIC criteria with ProtTest v2.4) [ 146].
The initial model determined by PHASER was adjusted with the program COOT [58] and refined using the program REFMAC5 [59].
