The 635 compounds that form the complete dataset used for this work were split into training (370 compounds), first validation (93) and external sets (used as the second validation set of 172 compounds) of the CAESAR model.
All study participants (n = 679) were used as the second validation group.
To better understand how the two training set expansion methods improve the predictions, we sub-sampled the gold-standard network at different sizes, and compared the performance of local modeling with and without training set expansion using the second mode of cross-validation.
We predicted the utility values in the validation sample using the second (full) random effects GLS mapping model to assess if the estimates were reliable.
During the cross-validation experiments using the second dataset, the training set (that consists of four fifths of samples in the same dataset) is simply used to build the classifier based on the preselected set of features.
The statistics from the validation using the first network shows a relative mean bias of about 1%, a relative mean absolute difference of 6%, and a mean absolute difference of 1.0 MJ m−2.
Using the first round of 10-fold cross validation as an example, the specific procedures of the undersampling technique are as shown in Figure 3. Based on the theory of 10-fold cross validation, type I, type II, type III, and type IV J-proteins are randomly divided into 10 equally sized parts, respectively.
The ridge regression parameter was optimized for each setting of the number of taps using the first fold of 10 fold cross-validation, and we report the accuracy on the remaining 9 folds.
While the published PvMYB4 data used for the first validation do not contain intermediate metabolite concentrations, a more recent study provides steady-state data for several of the pathway metabolites [ 20].
The second validation method involved the use of Harrell's C-statistic.
