GIST Develops AI Model for New Drug Candidate Exploration

Prediction of fusion regions in HoTS and Attention distribution of Transformer. Photo by GIST provided

[Asia Economy Honam Reporting Headquarters Reporter Cho Hyung-joo] The research team led by Professor Nam Ho-jung of the Department of Electrical, Electronics, and Computer Engineering at GIST (Gwangju Institute of Science and Technology, President Kim Ki-seon) announced on the 21st that they have developed an artificial intelligence technology that predicts the binding sites and interactions between drugs and target proteins based on protein sequences.

The model developed by the research team, HoTS, performs pre-training on the binding sites of drug-target proteins and then makes predictions, providing not only high prediction accuracy but also the rationale for drug-target protein interaction predictions. This enables researchers in new drug development to receive more reliable predictions of effective compounds.

This study extracted binding sites with compounds from a large-scale protein 3D structure database and trained a deep learning model based on CNN and transformer architectures to predict binding sites on protein sequences.

After learning the binding sites, the model can predict drug-target protein interactions through additional transformer layers based on that learning. As a result, the deep learning model can predict drug-target interactions along with the binding sites.

The HoTS model demonstrated higher predictive power than other deep learning models, and despite using only protein sequence information for binding site prediction, it showed performance comparable to other 3D structure-based prediction models.

Professor Nam Ho-jung stated, "This research achievement is a technology that greatly improves the efficiency of discovering effective compounds during the new drug development stage, and above all, it is significant in that it opens the possibility of new drug development for novel target proteins without 3D structural information." He added, "It is expected that this model will enable fast and efficient discovery of effective compounds during the drug development process in the future."

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