GIST Develops Technology to Predict Side Effects of Complex Prescription Drugs

[Asia Economy Honam Reporting Headquarters, Reporter Cho Hyung-joo] It is very common to be prescribed multiple drugs together for faster therapeutic effects. Since drug interactions can significantly influence each other, increasing the likelihood of side effects, caution is necessary.

On the 14th, Professor Nam Ho-jung's research team from the Department of Electrical Engineering and Computer Science at GIST (Gwangju Institute of Science and Technology) announced that they have developed an artificial intelligence technology that predicts side effects caused by drug-drug interactions based on gene expression data.

Considering that the main subjects of polypharmacy are patients and the elderly, predicting side effects caused by drug-drug interactions in advance is very important. However, the frequency of observing such side effects during the new drug development stage is low, making it difficult to detect side effects beforehand.

To address this issue, various computer-based side effect prediction models have been developed, but they cannot be applied to predict side effects between new drugs in the development stage and drugs already on the market. Additionally, these models have limitations in interpreting the mechanisms causing side effects.

The research team developed the DeSIDE-DDI artificial intelligence model that predicts drug-drug interactions based on drug-treated gene expression data.

Compared to existing related studies, it shows high prediction accuracy and can suggest genes related to drug interactions, enabling interpretation of the mechanisms behind side effect occurrences.

This study consists of a model that generates drug-treated gene expression data to predict side effects of various compounds in the drug development stage and a model that uses this data to predict drug-drug interactions.

The gene expression data generation model predicts gene expression data when cells are exposed to drugs based on the drug’s structure and attribute information, allowing side effect prediction even for drugs without actual experimental data.

Using the generated data, drug interaction prediction is performed. Within the AI model, a GLU (Gated Linear Unit) neural network unit is used to mimic simultaneous drug intake phenomena, ultimately enabling extraction of key genes.

Finally, entities (drugs) and relationships (side effects) are converted into a low-dimensional vector space to predict whether a side effect occurs between the drug pairs.

The trained DeSIDE-DDI model demonstrated high predictive performance, can be used to predict interactions for new drugs, and allows identification of highly relevant genes through weight analysis of the gene expression data used.

Professor Nam stated, “This research outcome is an important study that can be utilized as a drug safety monitoring system by predicting side effects caused by polypharmacy in advance.” Kim Eun-young, a student who participated as the first author, said, “It explains the mechanisms when side effects occur due to interactions, contributing to safety verification in the new drug development stage.”

This research conducted by Professor Nam’s team was supported by the “Development of Explainable AI-based Toxicity and Side Effect Prediction System for Drug Candidates” (Korea Research Foundation Mid-career Researcher Support Project) and the “Development of Virtual Human Drug Response Analysis System through Integration and Application of Project Experimental Data” (Ministry of Science and ICT Gene Donguibogam Project). It was published online on March 4 in the Journal of Cheminformatics.

This content was produced with the assistance of AI translation services.