Detecting Side Effects of Adverse Drug Reactions through Drug-Drug Interactions

Abstract

Adverse drug reactions (ADRs) from drug-drug interactions (DDIs) are a major public health and healthcare cost problem worldwide. Increasing complexity of treatments and ageing populations make managing ADRs more challenging. Standard methods for predicting these reactions do not exist as ADRs are often not detected until reported post-market by patients. Here we present a novel framework using Graph Neural Networks (GNNs) with self-supervised learning to model and predict ADRs from DDIs. By representing drugs as molecular graphs we capture spatial and chemical properties of drug interactions. This research contributes to pharmacovigilance by providing a robust framework to identify potential DDIs and support clinical decision-making.

Introduction

Adverse Drug Reactions (ADRs) are harmful or intolerable responses to medications that often require changes in treatment and can lead to serious health consequences and increased healthcare costs. Many ADRs occur post-marketing, with variability influenced by factors such as sex, geography, and healthcare access. Most ADRs are preventable, emphasizing the need for early prediction to improve patient safety, especially in developing countries.

This study aims to develop a predictive methodology for ADRs caused by drug-drug interactions using machine learning techniques including K-Nearest Neighbors (KNN), Decision Trees, and Graph Neural Networks (GNN). GNNs are particularly suited to model complex drug relationships using graph representations of chemical structures, aided by a self-supervised variational autoencoder to enhance learning and reduce overfitting.

The system workflow involves data preprocessing, feature extraction, and training multiple models to classify drug interactions and predict side effects. Evaluation metrics such as accuracy, precision, recall, and F1-score demonstrate that the proposed GNN and an extended CNN2D model outperform traditional methods, providing a promising approach to enhance ADR prediction and improve patient care.

Conclusion

In conclusion, this study highlights the critical need for an efficient and reliable method to predict Adverse Drug Reactions (ADR) due to drug-drug interactions, since ADR itself is a risk for public health. This new system utilizes a Graph Neural Network (GNN) based on self-supervised learning, predicting ADRs from drug interactions far more skillfully than existing methods. GNN successfully describes the drug-drug relationship to enhance its prediction and improve the incidence of unusual interactions. The model performs better than other existing algorithms and predicts ADRs with an accuracy of approximately 97.69%.

References

[1] J. Zhu, C. Che, H. Jiang, J. Xu, J. Yin, and Z. Zhong, ‘‘SSF-DDI: A deep learning method utilizing drug sequence and substructure features for drug–drug interaction prediction,’’ BMC Bioinf., vol. 25, no. 1, p. 39, Jan. 2024 [2] P. Bongini, F. Scarselli, M. Bianchini, G. M. Dimitri, N. Pancino, and P. Lió, ‘‘Modular multi–source prediction of drug side–effects with DruGNN,’’ IEEE/ACM Trans. Comput. Biol. Bioinf., vol. 20, no. 2, pp. 1211–1220, Mar. 2023. [3] F. Castiglione, C. Nardini, E. Onofri, M. Pedicini, and P. Tieri, ‘‘Explainable drug repurposing approach from biased random walks,’’ IEEE/ACM Trans. Comput. Biol. Bioinf., vol. 20, no. 2, pp. 1009–1019, Mar. 2023. [4] S. Abbas, G. Avelino Sampedro, M. Abisado, A. S. Almadhor, T.-H. Kim, and M. Mohamed Zaidi, ‘‘A novel drug-drug indicator dataset and ensemble stacking model for detection and classification of drug-drug interaction indicators,’’ IEEE Access, vol. 11, pp. 101525–101536, 2023. [5] J. Zhang and M. Xie, ‘‘NNDSVD-GRMF: A graph dual regularization matrix factorization method using non-negative initialization for predicting drug-target interactions,’’ IEEE Access, vol. 10, pp. 91235–91244, 2022. [6] C. Kim and N. Tatonetti, ‘‘Prediction of adverse drug reactions associated with drug-drug interactions using hierarchical classification,’’ bioRxiv, Feb. 2021. [7] C. Palleria, A. Di Paolo, C. Giofrè, C. Caglioti, G. Leuzzi, A. Siniscalchi, G. De Sarro, and L. Gallelli, ‘‘Pharmacokinetic drug-drug interaction and their implication in clinical management,’’ J. Res. Med. Sciences, vol. 18, no. 7, p. 601, 2013. [8] F. Del Pup and M. Atzori, ‘‘Applications of self-supervised learning to biomedical signals: A survey,’’ IEEE Access, vol. 11, pp. 144180–144203, 2023. [9] I. D. Mienye and Y. Sun, ‘‘A survey of ensemble learning: Concepts, algorithms, applications, and prospects,’’ IEEE Access, vol. 10, pp. 99129–99149, 2022. [10] P. Veli? kovi?, G. Cucurull, A. Casanova, A. Romero, P. Lió, and Y. Bengio, ‘‘Graph attention networks,’’ 2017, arXiv:1710.10903. [11] W. Hamilton, Z. Ying, and J. Leskovec, ‘‘Inductive representation learning on large graphs,’’ in Proc. Adv. Neural Inf. Process. Syst., 2017, pp. 1–19. [12] R. Jovanovic, M. Palk, S. Bayhan, and S. Voss, ‘‘Applying graph neural networks to the decision version of graph combinatorial optimization problems,’’ IEEE Access, vol. 11, pp. 38534–38547, 2023.

Copyright

Copyright © 2025 Kolla Lakshmi Harini, Rayanuthala Mercy, Jetti Siddhartha, Chitrala Yasaswi, Mrs. D. Srivalli. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.