Tri-Modal Molecular Representation Learning for ADMET Prediction Using a Multi-Task Framework
Abstract: Accurate prediction of absorption, distribution, metabolism, excretion and toxicity (ADMET) properties is essential in early-stage drug discovery. In this study, a tri-modal molecular representation learning framework is proposed to improve ADMET prediction performance. The proposed approach utilizes molecular descriptors, graph-based representations, and sequence information to improve ADMET prediction. It integrates fingerprint-based descriptors, graph neural networks, and transformer-based sequence approaches to capture complementary molecular information. A stacking ensemble strategy is employed to combine predictions from individual approaches into a unified output. The proposed approach is evaluated on six benchmark datasets, including BBBP, BACE, Caco-2, HIA, Tox21, and HIV. Experimental results demonstrate that the ensemble approach consistently outperforms individual models, achieving ROC-AUC scores ranging from 0.78 to 0.91. The multi-representation fusion enhances robustness and generalization across diverse molecular tasks. This approach supports medicinal chemists in prioritizing drug-like molecules by improving the prediction of pharmacokinetic and toxicity properties, thereby reducing experimental cost and time in drug development. The proposed framework achieved ROC-AUC values between 0.78 and 0.91 across benchmark datasets, outperforming individual representation models. The proposed tri-modal framework achieved competitive ROC-AUC performance across benchmark ADMET datasets, demonstrating the effectiveness of integrating fingerprint, graph, and sequence-based molecular representations for drug discovery applications.
Keywords: ADMET Prediction, Deep Learning, Graph Neural Networks, Multi-Task, Learning, Drug Discovery, Molecular Representation Learning
How to Cite?: Geetha D, Sathisha A D, Christy Binoy, Lokeshwari D M, Jagadish R, Anil Kumar R J, "Tri-Modal Molecular Representation Learning for ADMET Prediction Using a Multi-Task Framework", Volume 15 Issue 7, July 2026, International Journal of Science and Research (IJSR), Pages: 536-544, https://www.ijsr.net/getabstract.php?paperid=SR26706164720, DOI: https://dx.doi.org/10.21275/SR26706164720