Researchers in Korea have developed artificial intelligence (AI) that can design new drug candidates solely based on protein information. This is expected to reduce time and expense compared to traditional drug development while increasing the success rate.
The Korea Advanced Institute of Science and Technology (KAIST) stated on the 10th that a research team led by Professor Kim Woo-yeon from the Department of Chemistry developed an AI model called "Bind" that can simultaneously design and optimize optimal drug candidates and their binding mechanisms based solely on the structure of target proteins. This research achievement was published last month in the international journal "Advanced Science."
The key to this technology is "concurrent design." Existing AIs first create candidate substances and then separately evaluate whether they will bind to the target proteins. In contrast, Bind considers how candidate substances and proteins will bind together, allowing for simultaneous design. It is likely to create stable and effective candidates by reflecting important atomic positions and binding forms when binding with proteins.
Bind is designed to evenly meet important criteria in drug development, such as stability, physical properties, and naturalness of structure. Previously, it was common to meet only one or two conditions while failing to satisfy others, but this technology increased practicality by balancing multiple factors.
The operating principle is a "diffusion model" that gradually enhances completeness from a random structure. This is similar to the method utilized by "AlphaFold3," which won the 2024 Nobel Prize in Chemistry. However, while AlphaFold3 directly takes atomic coordinates, Bind incorporates criteria consistent with chemical laws, such as binding length or distance, to create more realistic structures.
The research team also applied an optimization method that reuses those molecules with outstanding binding patterns among those created. Through this, they produced superior candidates even without additional learning, successfully designing candidate substances that act specifically on certain mutations of the cancer target protein (EGFR).
Professor Kim stated, "Bind can independently learn and understand key elements that bind well to target proteins, allowing it to design optimal molecules even without prior information," adding, "Based on the principles of chemical interaction, it will enable faster and more precise drug development."
References
Advanced Science (2025), DOI: https://doi.org/10.1002/advs.202502702