Rapid Screening of Anticoagulation Compounds for Biological Target-Associated Adverse Effects Using a Deep-Learning Framework in the Management of Atrial Fibrillation.

Background: Deep learning methods may be useful for drug compound interaction and discovery analysis. However, there has been limited research on their use for screening biologically related adverse effects. Objectives: This study aims to pre-emptively screen for likely drug use persistence or success in clinical trials. Methods: This shall be achieved through the extension, application, and evaluation of a deep learning-based framework. Specifically, it shall be considered in the discovery of novel candidates and mechanisms underlying AF management-related adverse effects. The targets were linked to their adverse effects specified in two previous studies, their corresponding protein sequences, and the organs affected. Results: The new model showed good performance when compared to existing approaches in the Side Effect Resource (SIDER) and Food and Drug Administration Adverse Event Reporting System (FAERS) external validation datasets. A precision of 0.879 was obtained for enoxaparin, along with a recall of 0.746 for rivaroxaban. Stronger bleeding-related adverse effects were found for edoxaban compared with apixaban and enoxaparin. The binding and safety profiles of sequoiaflavone were very similar to those of rivaroxaban. Conclusions: This study presents a framework that could be used to pre-emptively screen for adverse effects. In doing so, it considers the biological basis for guiding optimal drug selection.