Support Vector Regression

We fed the structural features into a support vector regression model.

To avoid overfitting, we applied regularization to the regression model and estimated the out-of-sample performance on enantiomeric pairs that the model was not trained on.

The results for strictly using Morgan features were as follows:

Fig. 1 Test

With a correlation metric value of 0.5:

The results for strictly using Mordred features were as follows:

With a correlation metric value of 0.57: