Model comparison studies demonstrated that, after calibration, PF pesticide models clearly outperform chromatographic models for structured soils.
The resulting SP models clearly outperform empirical models in predictive ability outside the parameter space, as revealed by the leave-extremes-out cross-validation technique with omission of compounds beyond pre-defined lipophilicity and acidity ranges.
The FCFP_6 models clearly outperform the BCI models, which is not unexpected since the FCFP_6 fingerprints contain far more features than the BCI fingerprints and the FCFP_6 models were built using additional physical property descriptors.
With respect to the second question, adding hydrodynamic models to the model chain did, contrary to our expectations, not improve simulations, despite the fact that under idealized conditions (using observed instead of simulated lateral inflow) the hydrodynamic models clearly outperformed the routing schemes of the hydrological models.
In all cases, our models clearly outperform the naive predictor indicating that information on the whole antibody sequence effectively contributes to the prediction performance.
Overall, it can be seen that the local-score SCPE performs somewhat better than the global-score SCPE, and that both SCPE models clearly outperform JTT for a significant range of parameter ω around the optimum value.
As it makes more sense in coding sequences to assume a specific pattern of among-site rate variation for either the first and second codon positions grouped together ("+Γ112" in Figure 2) or for each codon position separately ("+Γ123" in Figure 3), the corresponding models clearly outperform a model which imposes 1 general pattern of among-site rate variation for all sites.
Furthermore, for LPXA, and for sites of the structurally constrained class i of all other families studied, both SCPE models clearly outperform the widely used JTT model, showing the power of the SCPE model to account for substitution patterns conditioned by structural constraints.
The signature gene expression models clearly outperform the randomly selected gene set for predicting disease mortality among ER positive patients (median C-index of 0.61 compared to 0.51), and do slightly better than the random gene set for predicting disease recurrence (median C-index of 0.56 vs. 0.52 for ER positive patients and 0.58 vs. 0.57 for ER negative patients).
The GEB model clearly outperforms the diffusionless T0 model.
