After the training is finished, the trained neural network is estimated for reproducing the target output values for the training data.
A QSAR model correlating calculated ΔΔGcomp with the measured ICexp50 values for the training set was prepared and confirmed that our computational approach can serve for reliable prediction of PR inhibitory potencies of peptidomimetics.
One issue for this approach would be obtaining the values for the training set.
Figure 2 shows the predicted values versus the photodegradation actual values for the training set.
Figure 6 presents the experimental and predicted values for the training and test sets for logP.
The predicted values computed using this MLR model with the experimental values for the training and test sets are shown in Table 4, and plotted in Fig. 2.
Table 4 shows the best model selected by RRegrs, its number of features, the adj.R2 and the R2 and RMSE values for the train and test sets, averaged over 10 random splits of the data.
In order to eliminate the influence of unavoidable outliers in training sample on a model's performance, a novel least square support vector machine regression, which combines outlier detection approach and adaptive weight value for the training sample, is proposed and named as adaptive weighted least square support vector machine regression (AWLS-SVM).
In an off-line evaluation of results, both models show good predicting abilities; in particular the ANFIS model presents an absolute error value for the training and test phases respectively equal to 0.7 and 0.9 (as a percentage of 3.5% and 4.5%), while, the same evaluation obtained using the ANN-BP model provides 0.92 and 0.9 (as a percentage of 4.6% and 4.5%).
We implement the learning adaptation function at training iteration i as (f(i) = (i+1)^{beta}), again with (beta= -0.15), a standard value for the training hyper-parameters.
The maximum SALI value for the training, test, and prospective data sets were approximately 200, 200 and 75, respectively, even though the prospective set had a slightly greater spread of Tanimoto values (Figure 4) Thus, the SALI summary values (SCI, S 0), S(1)) represent the SALI landscape covered by that data set only.
