Privileged Scaffold Analysis of Natural Products with Deep Learning-based Indication Prediction Model

Natural products play a vital role in the drug discovery and development process as an important source of reliable and novel lead structures. But the existing criteria for drug leads were usually developed for synthetic compounds and cannot be directly applied to identify lead scaffolds from natural products. To solve this problem, we propose a method to predict indications and identify privileged scaffolds of natural products for drug design. A deep learning model was built to predict indications for natural products. Entropy-based information metrics were used to identify the privileged scaffolds for each indication and a Privileged Scaffold Dataset (PSD) of natural products was constructed. The PSD could serve as a novel source of lead compounds and circumvent existing drug patents. This method could be generalized by replacing the training set, the prediction algorithm, and the compound set, to obtain more personalized-PSDs.