A Machine Learning Approach for Plant-based Drug Discovery: High-Throughput Prediction of Biological Activities and Enzyme Commission Numbers from Phytochemicals and Amino Acid Sequences of Plants

Abstract

The success of many plant-based drugs and the acknowledgement of the limitations of synthetic drugs has again sparked interest in plant-derived natural products (NP) as a valuable source for novel drug development. Researchers have traditionally used the knowledge-based approach, which relies on traditional medicines to identify candidate plants and extracts. However, NP-based drug development comes with many limitations during the screening stage. First, NP extracts are mostly incompatible with target-based or high-throughput screening. Furthermore, in the case of phenotypic assays, the deconvolution of the mechanism of action of the compound is costly and time-consuming. This study proposes a novel machine learning framework for the high-throughput identification and characterization of plant-derived NPs. This framework consists of two independent models. The first model is a neural network designed to predict phytochemicals’ bioactivities through multi-label classification in four categories: antioxidant activity, anti-inflammatory, neurotoxicity, and lipid metabolism. The second model is a convolutional neural network (CNN) that predicts the Enzyme Commission (EC) numbers of enzymes present in the plant. The proposed framework showed robust performance with the Bioactivity Prediction Model achieving 97.62% accuracy and the EC-number Prediction Model achieving 81.97% accuracy. The framework facilitates a more efficient NP-based drug development by providing important insights applicable to the screening, isolation, and deconvolution of NPs.