Machine Learning-Based Prediction of bioactive Compounds from medicinal plants for anticancer activity

Abstract

Medicinal plants remain one of the richest sources of anticancer therapeutics, with a substantial proportion of clinically approved oncology drugs — including paclitaxel, vincristine, vinblastine, and camptothecin derivatives — originating from plant secondary metabolites, yet the conventional pipeline for discovering and validating bioactive phytochemicals remains slow, costly, and heavily reliant on exhaustive wet-laboratory screening. This review consolidates the current literature and proposes a computational framework for machine learning (ML)-based prediction of bioactive anticancer compounds from medicinal plants. The framework spans data acquisition from phytochemical and bioactivity databases (ChEMBL, PubChem BioAssay, IMPPAT, NPASS, COCONUT, TCMSP), molecular representation (descriptors, fingerprints, and graph/SMILES-based embeddings), feature selection, and model development using Random Forest, Support Vector Machines, Gradient Boosting (XGBoost/LightGBM), Artificial Neural Networks, and Graph Neural Networks, benchmarked with accuracy, precision, recall, F1-score, ROC-AUC, and Matthews Correlation Coefficient. Synthesis of the published literature shows that ensemble tree-based models and graph-based deep learning architectures consistently outperform classical QSAR approaches, with accuracy and AUC scores exceeding 0.85 in multiple studies on plant-derived anticancer compound datasets, and that recurring pharmacophoric features — phenolic hydroxyl groups, nitrogen-containing alkaloid ring systems, and moderate lipophilicity — are repeatedly linked to cytotoxic activity. We conclude that data scarcity, class imbalance, limited interpretability, and a persistent gap between computational prediction and experimental validation remain the principal barriers to translation, and we outline multi-omics integration, explainable AI, and federated learning as priority directions for strengthening ML-guided, plant-based anticancer drug discovery.