Exploiting Hypergraph Topologies: Advancing Explainable AI for Predicting Drug Synergies

Background: Considering the importance of drug therapy as a dominant approach in cancer treatment, monotherapy has been successful in advancing disease treatments, but its effectiveness can be limited due to different drug responses. To overcome these challenges, the drug combination strategy involving the use of multiple drugs to treat a specific disease has been advocated. Objective: The objective of this research was to investigate the use of Hyper Graph Neural Networks (HGNNs) in modeling and predicting the interactions between drug combinations and their consequent effects on certain cell lines. Method: The methodology involved extensive data preprocessing, exploratory data analysis (EDA), and the implementation of an HGNN model tailored to capture the complex inter-relations of multidimensional data. The Ex-HGNN model showed superior performance metrics, including high accuracy, precision, recall, and F1-scores, and portrayed efficiency in categorizing drug interactions and their effects as either synergistic or antagonistic. A critical detail of this study was the implementation of explainable AI approaches, such as SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations). These methods brought transparency into the decisions of the Ex-HGNN model, rendering them more interpretable and trustworthy. They allowed insight into the contribution of individual features on the prediction decisions of the Ex-HGNN model, an essential component in the field of drug efficiency analysis. Results: Experimental results validated the proposed drug synergy prediction model and its significant enhancement compared to state-of-the-art methods. Conclusion: These findings provide a stepping-stone for future research using machine learning and deep learning approaches in resolving other drug-related issues. This work demonstrates the effectiveness of Hyper Graph neural network modeling for high-dimensional data analysis and emphasizes the importance of explainable artificial intelligence in the fields of healthcare and precision medicine.