Leveraging Deep Learning to Uncover Hidden Patterns of Adverse Events in FAERS Data

This project aims to use the FDA Adverse Event Reporting System (FAERS) to identify safety signals for newly marketed drugs by applying advanced AI and machine learning methods. FAERS is a widely used public source of adverse event reports for most drugs. This project will focus on identifying subgroups of patients who have similar side effects from medication and the most frequent adverse event profiles within those subgroups. We will leverage deep learning via large language models and consensus clustering-based methods to identify patients with similar side effects from a medication. On the one hand, the goal is to group semantically similar patient profiles despite heterogeneity of terminology used in textual symptom descriptions or drug names due to multiplicity of factors, such as demographics. On the other hand, we also aim to identify the most frequent adverse event profiles within the extracted subgroups.