Korea University Study Uses Artificial Intelligence to Design Potential Antiviral Candidates

Researchers from Korea University have developed a new artificial intelligence-supported strategy targeting host cell proteins to combat viral infections. The study addressed the inhibition of the human enzyme deoxyhypusine synthase (DHPS) to overcome drug resistance in viruses.

The rapid evolution of viruses and their resistance to drugs developed directly against them make the treatment of infections difficult. Seeking a solution to this problem, a Korea University team led by Professor Kwang Yeon Hwang from the Faculty of Life Sciences and Biotechnology and Associate Professor Hong-Rae Kim from the College of Medicine focused on a host-targeted therapeutic approach focusing on host proteins. The researchers developed an artificial intelligence-supported strategy to discover molecules that block DHPS, a human enzyme in the eIF5A hypusination process that supports the replication of a wide range of viruses. As part of the study published in the Journal of Cheminformatics on August 5, 2026, 13 machine learning models to predict molecular binding scores and 10 methods to generate new molecules were systematically compared.

Among the tested models, Out-of-Fold Stacking showed the best overall performance, while RL-Design achieved the highest predicted hit rate among generation methods at 92.02%, and 98.0% of top-tier molecules passed the standard drug-likeness threshold. To experimentally evaluate the AI-generated molecules, six fragments were selected, predicted as three active and three inactive; all predicted active fragments inhibited DHPS while the inactive ones showed no effect. Two of the active fragments offered measurable IC50 values of 31.2 and 49.1 micromolar. Furthermore, X-ray co-crystal structure provided structural evidence for the binding of one of the AI-derived compounds at the entrance of the DHPS substrate and NAD-binding pocket.