Abstract
(1) Background: Although the field of natural product (NP) drug discovery has been extensively developed, there are still several bottlenecks hindering the development of drugs from NPs. The PD-1/PD-L1 immune checkpoint axis plays a crucial role in immune response regulation. Therefore, drugs targeting this axis can disrupt the interaction and enable immune cells to continue setting up a response against the cancer cells. (2) Methods: We have explored the immuno-oncological activity of NPs targeting the PD-1/PD-L1 immune checkpoint by estimating the half maximal inhibitory concentration (IC(50)) through molecular docking scores and predicting it using machine learning (ML) models. The LightGBM (Light Gradient-Boosted Machine), a tree-based ML technique, emerged as the most effective approach and was used for building the quantitative structure-activity relationship (QSAR) classification model. (3) Conclusions: The model incorporating 570 spectral descriptors from NMR SPINUS was selected for the optimization process, and this approach yielded results for the external test set with a sensitivity of 0.74, specificity of 0.81, overall predictive accuracy of 0.78, and Matthews correlation coefficient (MCC) of 0.55. The strategy used here for estimating the IC(50) from docking scores and predicting it through ML models appears to be a promising approach for pure compounds. Nevertheless, further optimization is indicated, particularly through the simulation of the spectra of mixtures by combining the spectra of individual compounds.