Review on Explainable Modeling of Drug–Drug Interactions Using Heterogeneous Graph Structures
TL;DR
A full overview of AI-based DDI prediction, including information like, but not limited to, publicly available resources for AI-DDI study, approaches for data handling and feature preprocessing, explainable Artificial Intelligence (XAI) schemes that improve trust toward an approach based on other XAI methods that contribute to achieving faith in the ability of DDI prediction method since it falls under the critical tasks category and modeling methodologies are discussed.
Abstract
Over the past decade, polypharmacy has become increasingly prevalent in therapy. The unwanted DDIs with unintended side-effects that result from the interoperability of heterogeneous treatment regimens are still a major concern. The wide application of AI technologies has led to the development of various AI prediction models for predicting DDI to facilitate drug selection for physicians. However, the opacity of AI models raises questions about their reliability and these models definitely have great potentials to be harnessed for aiding physicians in polypharmacy decision-making tasks. The acute problem mentioned above can be listened to in explaining the methodology for adding layers of AI Models. Explainable AI (XAI) encourages safety and transparency by outlining how predictions are formed in a model used for prediction, such as our work on DDI forecast. The review includes a full overview of AI-based DDI prediction, including information like, but not limited to: publicly available resources for AI-DDI study, approaches for data handling and feature preprocessing, explainable Artificial Intelligence (XAI) schemes that improve trust toward an approach based on other XAI methods that contribute to achieving faith in the ability of DDI prediction method since it falls under the critical tasks category and modeling methodologies. Finally, we discuss XAI limitations and possible future developments in DDIs.
