A review on graph neural networks for predicting synergistic drug combinations

Artificial Intelligence Review (2024) 57:49 https://doi.org/10.1007/s10462-023-10669-z A review on graph neural networks for predicting synergistic drug combinations Milad Besharatifard4 · Fatemeh Vafaee1,2,3,4 Accepted: 20 December 2023 / Published online: 13 February 2024 © The Author(s) 2024 Abstract Combinational therapies with synergistic effects provide a powerful treatment strategy for tackling complex diseases, particularly malignancies. Discovering these synergistic combinations, often involving various compounds and structures, necessitates exploring a vast array of compound pairings. However, practical constraints such as cost, feasibility, and complexity hinder exhaustive in vivo and in vitro experimentation. In recent years, machine learning methods have made significant inroads in pharmacology. Among these, Graph Neural Networks (GNNs) have gained increasing attention in drug discovery due to their ability to represent complex molecular structures as networks, capture vital structural information, and seamlessly handle diverse data types. This review aims to provide a comprehensive overview of various GNN models developed for predicting effective drug combinations, examining the limitations and strengths of different models, and comparing their predictive performance. Additionally, we discuss the datasets used for drug synergism prediction and the extraction of drug-related information as predictive features. By summarizing the state-of-the-art GNN-driven drug combination prediction, this review aims to offer valuable insights into the promising field of computational pharmacotherapy. Keywords Graph neural networks · Drug combination · Synergy prediction · Cancer treatment 1 Introduction Combination therapy, a treatment modality that combines two or more therapeutic agents, has increasingly become the preferred approach for many human diseases, especially those caused by alterations in multiple genes or pathways, such as cancer. The integration * Fatemeh Vafaee 1 School of Biotechnology and Biomolecular Sciences, University of New South Wales (UNSW), Sydney, Australia 2 UNSW Data Science Hub, University of New South Wales (UNSW), Sydney, Australia 3 OmniOmics Pty Ltd, Sydney, Australia 4 Biomedical AI Laboratory (Vafaee Lab), Sydney, Australia 13 Vol.:(0123456789) 49 Page 2 of 38 M. Besharatifard, F. Vafaee of anti-cancer drugs enhances efficacy compared to using a single therapy, as it targets different key pathways in a synergistic or additive manner. By combining drugs with distinct mechanisms of action, therapeutic effectiveness can be enhanced, allowing for lower-dose prescriptions, and reducing the potential risks of side effects and toxicity. Clinical evidence consistently demonstrates the utility of combining different therapeutics to improve treatment efficacy in various cancer types, such as breast cancer (Fisusi and Akala 2019), lung cancer (Molina-Arcas et al. 2019), and ovarian cancer (Lui et al. 2020), among others. However, the search for effective combinations is hindered by the sheer number of potential drug pairs, leading to a combinatorial explosion (Azad et al. 2021b; Gilvary et al. 2019). It is infeasible to experimentally screen the enormous search space of all possible drug combinations. Consequently, the development of computational models to identify potential anti-cancer synergistic drug combinations efficiently and accurately has garnered significant attention from both the scientific community and the pharmaceutical industry. With the increasing availability of large-scale high-throughput screening datasets for identifying synergistic drug combinations, a growing number of artificial intelligence (AI) methods are being employed for in silico predictions of efficacious drug combinations (Hosseini and Zhou 2023; Hu et al. 2022; Zhang et al. 2023; Zhang and Tu 2023; Wang et al. 2022a). Among different AI models, GNNs have emerged as a powerful class of artificial neural networks designed to process and learn from data structured as graphs. Graphs consist of nodes (vertices) connected by edges (links or relationships), and they are widely used to represent complex relationships and interactions between different entities. Due to their versatility, GNNs have found applications in various fields, including computer vision, natural language processing, social network analysis, bioinformatics, and drug discovery, among others (Zhou et al. 2020). The increasing importance and application of AI and machine learning in drug discovery have prompted different review articles outlining various data sets, machine learning algorithms, and deep learning models developed to predict synergistic drug combinations in cancer (Torkamannia et al. 2022; Wu et al. 2022; Pearson et al. 2023; Kumar and Dogra 2022). For instance, Torkamannia et al. (Torkamannia et al. 2022) comprehensively reviewed a wide array of drug development data sources, encompassing biological datasets like molecular omics data, drug target information, and molecular interactions, as well as datasets containing high-throughput in vitro screening of drug combinations. Additionally, they presented an overview of the literature on computational methods designed for drug synergy prediction, broadly categorized into deep learning (DL), traditional machine learning (ML), and network-based methods. Around the same time, Wu et al. (2022)- performed a similar review of machine learning methods used in drug combination prediction across algorithmic categories of systems biology or network-based methods, kinetic models, mathematical models, stochastic search algorithms, classic machine learning, and deep learning methods. They summarized 29 studies, providing details of their respective algorithms, drug combination datasets, input data types, and the availability of program code. Further, Kumar et al. (2022) conducted a review focused on deep learning-based techniques for the prediction of synergistic drug combinations in cancer. They performed a comparative analysis of prediction techniques based on various performance measures. Additionally, they covered the theoretical aspects of drug synergy and scoring models at length with their mathematical formulations. However, all these reviews were conducted before the surge of GNN techniques in drug discovery. Therefore, they neither adequately cover GNN-related drug combination prediction studies nor represent recent advancements in GNN algorithms. 13 A review on graph neural networks for predicting synergistic… Page 3 of 38 49 The rise in the use of GNNs in drug discovery is due to their ability to handle and interpret complex data, such as molecular graphs and biological networks (Bongini et al. 2021; Zhao et al. 2021). GNN-based models have demonstrated high performance and have yielded promising results in various aspects of drug discovery, including virtual screening, molecular property prediction, protein–ligand bindi (...truncated)