The Adverse Medication Interaction and Disease Correlation Analysis Using Deep Learning

Medication interaction refers to the alteration of the action of one medicinal agent by virtue of the co-administration of another substance, which may include another drug, food products, or a specific pathological condition. Such interactions have the potential to cause severe pharmacological effects, highlighting the need for a quick, computational approach to predict medication interactions, as conventional methods are often time-consuming. To address this, a Large Language Model (LLM), trained on a vast corpus of data, was chosen as the base model and fine-tuned with a benchmark medication dataset to predict interactions between medications. Parameter-efficient fine-tuning techniques, including QLoRA, were applied to optimize performance while conserving computational resources. The resulting framework captures drug characteristics and interaction dynamics, ultimately predicting both the occurrence of a reaction and the expected outcome. For example, the combination of Diclofenac and Lepirudin significantly increases the risk of bleeding and hemorrhage. To a great extent, the computational models developed to predict interactions of drugs act as a good tool, which provide useful information for combination therapies and helping clinicians make more well-informed decisions, ultimately improving treatment outcomes and lowering the risk of adverse effects.