[Exclusive] AI Drug Development Breakthrough: Achieving Both Efficacy and Productivity

Joint Research Team Led by Professors Baek Min-kyung and Yoon Tae-young at Seoul National University

Application of AlphaFold and ProteinMPNN on the SPID Platform

Enhancing Both Productivity and Binding Affinity


Development of Ada

Korean researchers have taken a significant step forward in the development of artificial intelligence (AI)-powered antibody drugs that achieve both productivity and binding affinity. They have overcome the longstanding limitation of traditional antibody engineering, in which increasing the binding affinity of AI-designed antibodies came at the expense of productivity, thereby demonstrating the practical effectiveness of AI-based drug design.

According to the pharmaceutical, biotech, and academic sectors on the 27th, a joint research team led by Professors Baek Minkyeong and Yoon Taeyoung from the Department of Biological Sciences at Seoul National University published their findings on the international preprint platform 'bioRxiv.' The paper, entitled "Structural Logic of AI-Guided Antibody Rescue and Therapeutic Optimization via Large-Scale Data Landscape Exploration," has been released in preprint form on bioRxiv ahead of formal publication in an international academic journal.

AI antibody drugs are considered a new means of advancement in the pharmaceutical and biotech industries, yet the dilemma between binding affinity and productivity has long been a chronic challenge during their development. Even when an antibody with strong affinity for a disease-causing antigen is discovered, attempts to mass-produce it in a pharmaceutical plant often result in a sharp decline in productivity. This is due to the phenomenon of 'negative epistasis,' where amino acids with excellent individual properties, when combined into a single antibody, actually undermine binding affinity or severely damage productivity. As a result, drug development experts have pointed out that AI may not be suitable for antibody drug development.

The root cause of this phenomenon lies in the instability of the antibody protein, which occurs when the amino acid sequence of the complementarity determining region (CDR)—the key site where antigens and antibodies bind—fails to harmonize with the overall antibody framework. If the CDR sequence is modified to enhance binding affinity and this segment physically clashes with the rest of the antibody's three-dimensional structure, the antibody protein cannot fold correctly and its structure is compromised. Cells autonomously discard these structurally defective proteins, making it impossible to mass-produce the antibody in a factory setting.

The research team utilized the single-molecule protein interaction detection (SPID) platform owned by the Korean biotech company Proteina, which enables precise analysis of tens of thousands of antibody variants per week. Through this platform, they obtained data on 9,517 variants of adalimumab (an autoimmune disease treatment marketed as Humira), measuring both binding affinity and productivity. Their findings revealed that these two properties are intricately intertwined, like a rugged and irregular mountain range, showing that simply combining favorable mutations is not sufficient to create an optimal antibody.

To address this complexity, the researchers applied AI technologies. Specifically, they used the protein structure prediction AI 'AlphaFold3' and the inverse protein design AI model 'ProteinMPNN' to uncover hidden rules governing the three-dimensional structure and productivity of antibodies.

This allowed the AI to analyze the structure of antibodies that exhibited strong binding affinity but low productivity—candidates at risk of being eliminated as drug candidates—and suggest a customized solution: replacing just a single amino acid. When the research team applied this structure-based 'rescue strategy' to actual antibodies, they succeeded in restoring cell culture productivity to normal levels while maintaining the antibodies' binding affinity to the target.

The final optimal next-generation adalimumab variants discovered (clones 1207 and 1208) demonstrated up to a 100-fold improvement in in vivo efficacy compared to existing therapies in animal experiments. When these new antibodies were administered to mouse models with artificially induced psoriasis, only 1/20th to as little as 1/100th of the standard treatment dose was needed to normalize severe skin inflammation and thickness. The results proved a 100-fold increase in therapeutic efficacy by suppressing only the disease, without side effects or weight loss.

This near-miraculous improvement over conventional drugs was found to be due not only to the antibodies' stronger binding to their targets, but also to a dramatic increase in 'complex lifetime'—the duration that the antibody remains bound after attachment. Measurements showed that while conventional adalimumab maintained binding for about 1.8 hours, the newly designed antibody variants sustained their binding for as long as 50 hours, greatly extending their therapeutic effect.

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