Hanyang University develops AI to precisely predict 3D protein structures, cutting drug development time and costs

[Financial News] Hanyang University said on the 12th that a team led by Professor Kim Eun-sol of its Department of Artificial Intelligence had developed artificial intelligence (AI) technology that precisely predicts and generates the three-dimensional structures of proteins, a key step in drug development.
The research is expected to help dramatically reduce the time and cost involved in identifying drug candidates.
Proteins do not have fixed shapes; they act by flexibly changing shape depending on their surrounding environment or the molecules they bind to. For this reason, accurately analyzing the various three-dimensional states a protein can adopt is essential when developing new drugs. However, existing AI generative models had to produce a protein’s overall shape and detailed structure all at once. This created a problem: making the shape stable reduced diversity, while increasing diversity led to structural errors.
The research team applied a new approach that generates protein structures sequentially at multiple levels of detail. First, it establishes the protein’s overall backbone and framework, then uses that as a basis to progressively refine the detailed geometric structure.
This technology makes it possible to accurately predict the various changes in shape that occur when a protein interacts with a drug while maintaining the protein’s own stable form.
In tests against existing single-resolution generative models, the AI technology developed in this study significantly reduced errors in predicting flexible protein structures. It also maintained high accuracy without distorting the overall shape, even for proteins with long and complex structures.
Professor Kim Eun-sol said, “By separating a protein’s overall framework from its detailed structure and generating them in stages, we can achieve both structural stability and diversity of shape.” Kim added, “I expect it to serve as a practical modeling foundation in life sciences and drug development, fields where multiple states of proteins need to be considered together.”
The research, conducted with support from the Ministry of Science and ICT and the Institute for Information & Communications Technology Planning & Evaluation, was published in the international journal Bioinformatics.  

[email protected] Kim Man-gi Reporter