For each cross-validation calculation, this procedure created a partial training data set that contains response to combinations of 11 drugs and two different doses of a single drug totaling to 78 unique conditions (Supplementary file 1B,C).
We then performed a 10-fold cross-validation calculation to prune the full tree in order to minimise the error rate when generalised to unseen observations, and converted it into a classification mosaic chart.
For each cross-validation calculation, the response profile to every possible combination of an individual drug is withheld and models are constructed using the partial data (i.e., combinations of all other drugs plus the drug of interest applied as a single agent) and prior information.
As shown by cross-validation calculations, use of prior information significantly improved the predictive power of the models.
(See Figure 3 figure supplement 1 for the cross-validation calculations with all other partial data sets and Supplementary file 1E for statistical validation of the predictions).
Ten-fold cross-validation and calculation of the area under the curve (AUC) of the receiver operating characteristic (ROC) curve were used to evaluate model performance.
Figure 1 provides the 2 × 2 contingency table demonstrating the results of the leave-one-out cross validation procedures, and the associated performance calculations.
The calculation of internal consistency and cross validation provided evidence of reliability and lack of redundancy of items.
We used PROC GAM in SAS® (SAS Institute, Inc., Cary, North Carolina, USA), specifying automated calculation of df using the generalized cross validation method and loess smoothing, to assign df to response curves depicting calving season movement patterns in each elk.
The calculation of internal consistency, reproducibility, and cross validation provided evidence of reliability and lack of redundancy.
