Proteins are involved in all biological phenomena in the human body, from enzymes that act as catalysts in biochemical reactions to antibodies that fight bacteria and viruses, and hormones such as insulin that regulate metabolism. When there are problems with proteins, diseases can occur. Abnormal folding or aggregation of proteins can lead to Alzheimer's disease and Parkinson's disease, while excessive activation of certain proteins can cause cancer.
A domestic bio corporation has developed an AI capable of predicting the structure of proteins and designing proteins that could become new drugs. Currently, many companies use AI to find new drug candidates, but most utilize a method that identifies the optimal structures from protein databases to improve performance. Galux, founded by Professor Seok Cha-ok of Seoul National University, successfully designed completely new proteins using AI.
Galux has designed six types of antibody proteins that can be used for disease treatment. This method creates new proteins from scratch without referencing existing proteins and is called 'de novo' protein design, which means 'new and unlike anything existing' in Latin. It is considered the most innovative yet challenging technology in the field of drug development.
◇Six new types of antibodies that never existed before unveiled
In an interview at Galux's headquarters in Gwanak-gu, Seoul, CEO Seok said, "De novo technology is about designing molecules that never existed by not using information from already known substances," adding, "By replacing what has been done through computational chemistry with AI, it becomes possible to design new proteins once AI fully understands the principles of protein structure."
Proteins are forms of 20 different amino acids connected in various ways. DNA genes carry information that specifies the order of amino acid connections. While understanding DNA might suggest that proteins can be easily comprehended, the reality is different. The process by which amino acid chains fold into three-dimensional structures has so many variables that predicting their conformations based solely on genetic information is challenging.
AI has brought revolutionary advancements in protein structure interpretation. AI has learned from the information of proteins whose structures have been identified so far and has independently established the relationship between genetic information and the three-dimensional structures of proteins. A path has opened for understanding protein structures without conducting experiments. Galux has taken this further by designing proteins that do not currently exist without learning from existing protein structure interpretation data.
On the 17th, Galux unveiled six novel antibody candidates designed by the protein design AI, 'Galux Design,' on the preprint site bioRxiv. The antibodies bind to the PD-L1, HER2, EGFR, ACVR2A/B, FZD7, and ALK7 proteins to attack pathogens.
Galux announced that it has verified the actual binding strength and stability of the antibodies, proving that they are comparable to or superior to currently commercialized antibody drugs. Notably, ALK7's structure has yet to be revealed, but a new antibody tailored for it has been designed. This is akin to developing a custom missile without knowing who the enemy is.
Only three places in the world, including Galux, have succeeded in de novo antibody design. Last year, the U.S. company Nabla Bio disclosed a case of antibody design via de novo, while in 2021, David Baker, a Nobel Prize winner in chemistry, announced two discoveries through the AI de novo platform, RosettaFold.
Galux has successfully designed the most diverse antibody proteins revealed to date. Moreover, it has also enhanced binding strength compared to antibodies previously disclosed by Nabla Bio and Professor Baker. They developed an antibody that binds at concentrations of nanomoles (nmol), which is one billionth of a mole (mol), yet the strongest antibody found by Galux binds even at concentrations of picomoles (pmol), which is one trillionth of a mole, making it a thousand times smaller than nanomoles.
CEO Seok explained that the smaller the size, the higher the design precision, allowing precise action on target proteins. He stated, "Our body has many similar proteins, so a drug targeting protein A may not only act on A, but also on A′, A″, or even completely different locations. We have increased precision and sensitivity to ensure that it precisely targets only protein A, distinguishing fine differences between A and A′, which is Galux Design's distinguishing feature."
◇Applicable to ADC, dual antibodies, and vaccines
CEO Seok pointed out the differentiation in the AI learning approach as a key factor in Galux's de novo research success. While the existing AI firms' approach involves feeding a vast amount of data to memorize and learn protein structure patterns, Galux adopted a strategy of teaching the principles of protein structures from the beginning, enabling AI to fully understand how proteins function and how they are arranged.
He explained, "When solving math problems, memorizing similar types can boost problem-solving capabilities for that type, but when encountering more complex or unfamiliar problems, additional learning is required. In contrast, thoroughly understanding the principles enables the AI platform to not only solve various types but also to generate entirely new problems."
He added, "Even with the same data, the AI produces completely different results depending on how it is presented and how learning is induced. Galux created an AI that understands the principles of proteins from the ground up, rather than modifying or referencing existing data."
The success of de novo design at Galux can be attributed to CEO Seok's know-how. He established it in 2020 with his students from the Department of Chemistry at Seoul National University. Seok has been a scientist researching protein structure prediction for over 20 years based on physical chemistry. He received recognition in academia and served as a judge for the International Protein Structure Prediction Competition (CASP) in 2018 and 2020. The AI protein structure prediction technologies AlphaFold 1 and 2 from Google DeepMind were announced at that competition.
Through this research, Galux confirmed the high potential for Galux Design to apply to various modalities, including antibody-drug conjugates (ADCs), dual antibodies, cell therapies, and vaccines. CEO Seok stated, "Galux's AI platform is not limited to specific modalities and can be utilized for various diseases, including cancer, metabolic diseases, and brain diseases," adding, "Through tests conducted over the past two years, we have confirmed the possibility of increasing the efficacy and success rates of new drugs by enhancing precision and sensitivity."
Moving forward, Galux plans to expand its business based on de novo design. It is currently collaborating with domestic new drug development companies, including LG CHEM and Y-Biologics, to jointly develop new drugs, while also pursuing global partnerships. CEO Seok noted, "Now that we have completed the predictive and design platform, our goal is to create cases that lead to new drug development," and mentioned that discussions are underway with several global pharmaceutical companies.
References
bioRxiv (2025), DOI: https://doi.org/10.1101/2025.03.09.642274