[K-Bio on the AI Testbed]③Proteina Seeks Breakthrough in Overcoming the Biggest Challenge in Drug Design

AI-Driven Loop: Protein Prediction, Experimentation, and Design
AI Uncovers CDR Rules, Boosting Proteina's Competitive Edge

Editor's NoteArtificial intelligence (AI) is fundamentally reshaping the landscape of the pharmaceutical and biotech markets. NVIDIA's decision to collaborate with global pharma giant Eli Lilly, investing 1 billion dollars (approximately 1.499 trillion won) to pursue AI-driven drug development, symbolizes a new paradigm that is rewriting the history of the pharmaceutical and biotech industries. The domain of AI, once limited to candidate molecule discovery, is now expanding across the entire drug development cycle, including sophisticated clinical design, optimal patient group selection, and real-time manufacturing process control, radically restructuring the economics of the industry itself. AI has shortened the average timeframe for candidate identification from 3–4 years to as little as one year, and leveraged its 3D understanding of target proteins and improved toxicity prediction accuracy to raise Phase I clinical success rates from the typical 40–65% up to 80–90%. AI's effectiveness in reducing the risk of costly early-stage failures provides a crucial breakthrough for companies facing the 'valley of death' in drug development. Through the special series 'K-Bio on the AI Testbed,' The Asia Business Daily examines the high-speed reconfiguration of the global new drug development scene in 2026 and explores the paths for survival and advancement that Korea's bioindustry must pursue.

The axis of competition in Korea’s AI-driven drug development is split between 'data observation' and 'protein design.' The Samsung-Proteina alliance has adopted a strategy centered on 'analytical AI,' leveraging the accumulation of large-scale experimental data to uncover antibody rules. In contrast, Galux, collaborating with Celltrion, takes a 'generative AI' approach, designing protein structures that do not exist in nature. These two camps stand at the very forefront of the global AI drug development battlefield. We spoke in turn with both companies, each of which is redrawing the map of Korea's AI-powered drug industry from the midst of this intense competition.

<2> Proteina and Professor Paik Minkyoung's Team Discover the Elusive CDR Rule

In AI-driven drug development, 'speed' is emerging as the key variable for survival. As the traditional approach, which takes years to identify candidate molecules, can no longer keep pace with global competition, domestic and international biotechs are turning to AI strategies to engage in a race for speed.

Yoon Taeyoung, CEO of Proteina, is being interviewed at the Proteina headquarters in Guro-gu, Seoul. Photo by Jeong Donghoon

At the center of this effort, Proteina, an AI antibody drug developer, and Professor Paik Minkyoung’s research team at Seoul National University, have achieved a meaningful breakthrough. They have uncovered the rules of the CDR (complementarity-determining region), which had been considered the greatest challenge in AI-driven antibody drug development. These six loop structures, which directly bind to antigens, have highly variable and 'flexible' configurations, making them difficult for even AI to design. Professor Paik's antibody design AI, 'AbGPT-3D,' discovered these rules and further verified that they are applicable to real-world antibody drug AI development.

Yoon Taeyoung, CEO of Proteina (and Professor in the Department of Biological Sciences at Seoul National University), said in an interview with The Asia Business Daily on March 9, "Because these conditions were identified from nearly hundreds of thousands of data points, I am internally convinced that these rules apply broadly across antibodies."

"There Were CDR Rules"... Reliability Verified Through Data

The core platform of Proteina is 'SPID (Single-molecule Protein Interaction Detection).' This technology quantifies protein-protein interactions (PPI) at the single-molecule level and formed the foundation of their clinical sample analysis business when the company was established in 2015. Today, the platform has evolved to incorporate AI, enabling rapid and compressed development of antibody drug candidates. Proteina’s approach can be summarized as 'Analytical AI'—'seeing what is there with precision.' They precisely measure actual antibody behavior through large-scale experimental data and feed those results back into the AI design process. Professor Paik’s team trains AbGPT-3D under various conditions, while Proteina compares these results against massive experimental data generated by its SPID platform to identify the best matching conditions.

CEO Yoon used a satellite imagery analogy, saying, "Traditional antibody research was either like taking a black-and-white photo of the entire city of Seoul or shooting just one block of the Guro Digital Complex in ultra-high resolution. Our technology enables us to capture an area 100 to 1,000 times larger than before, while maintaining high resolution, and it was this large-scale perspective that allowed us to discern the underlying rules."

These rules are not merely an academic discovery. They represent training conditions that can actually be applied in generative AI antibody design. The reliability has been verified in practice: when these conditions were applied to their accumulated internal data, they found a suitability rate exceeding 90%.

From 5,000 to 30,000 Per Week... 'Speed Innovation' Driven by Samsung Partnership

The discovery of CDR rules was made possible by a technological leap in the SPID platform. Last fall, Proteina’s antibody sequence processing capacity was about 5,000 per week; now, it has increased to 15,000–30,000 per week. This was achieved through both equipment automation and scale-up, in response to the demand for large-scale data production during collaboration with Samsung Bioepis. CEO Yoon noted, "Our team even gave up weekends and holidays to optimize the entire process."

Yoon Taeyoung, CEO of Proteena, is explaining the company's AI drug development platform, SPID, at Proteena's headquarters in Guro-gu, Seoul. Photo by Jeong Donghoon

Proteina has formed a consortium with Samsung Bioepis and Professor Paik’s team to execute a government-funded project worth 47 billion won, focused on 'AI-powered antibody drug development.' The goal is to discover 10 new drug candidates and submit one IND (Investigational New Drug application) by 2027. The supply of candidate molecules for Samsung’s designated targets is proceeding smoothly, and there have been no major setbacks in the schedule so far.

The reason Samsung chose Proteina lies in this 'experimental big data production capability.' To select the highest-quality candidates from thousands or tens of thousands of AI-designed molecules that are suitable as real drugs, large-scale experimental validation is required. Proteina’s next goal is to handle 100,000 candidates per week, but they believe this will require not just scale-up, but innovation in the platform itself. Since Proteina manufactures its own chips, equipment, and software, development is underway with the goal of achieving this next year.

From 'Best-in-Class' to 'First-in-Class'... De Novo Design Remains the Final Challenge

Proteina’s current strategy is 'compressed development of biobetters.' Starting with existing antibodies that have already demonstrated efficacy in preclinical and Phase I trials, they use a closed-loop system of AI and autonomous experimentation to produce new candidates with enhanced efficacy and stability in just 3–4 months. Since these are completely redesigned sequences, they are free from existing patents. In terms of drug development strategy, this is 'best-in-class.' The next goal is 'de novo'—to design antibodies entirely from scratch, or in other words, to achieve a 'first-in-class' breakthrough.

CEO Yoon said, "No company in the world has achieved this yet, but looking at current trends, I believe it must be realized within the next one or two years." He explained that the discovery of CDR rules narrows the scope of random exploration and that the accelerated verification process supports this ambitious goal.

In the first half of this year, Proteina will release three internally developed assets: a small-molecule drug for degenerative arthritis, an anti-obesity antibody, and a bispecific antibody for autoimmune diseases. The anti-obesity antibody is positioned to prevent rebound weight gain after GLP-1 class treatment. Their work on GPCR (cell membrane protein) targets—considered the most challenging in antibody development—has achieved low-dose, long-acting properties, drawing attention in academic circles and recently being selected for oral presentation at the U.S. Obesity Society. The official announcement is scheduled for early June 2026. CEO Yoon’s keyword for this year is 'demonstration.' Proteina’s goals are to enter one or two phase I clinical trials by next year, to add two to three new assets every six months, and to achieve visible results in alliances with global pharmaceutical companies in the first half of next year.

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