"A DeepSeek Moment has come to bio AI."
China's ByteDance Ltd., which created the social media platform TikTok, has unveiled a protein structure prediction artificial intelligence (AI) model, Protenix, sending ripples through the drug development market.
After Google DeepMind's AlphaFold opened the era of AI-based drug development, U.S. and Chinese technology corporations, including Nvidia, Insilico Medicine, and ByteDance Ltd., are rushing in.
In Korea, Galux, which has de novo technology to design proteins from scratch using AI, and Proteina(468530), which develops new drugs and cancer companion diagnostics by combining big data on protein-protein interaction (PPI) with AI, have commercialized their technologies and entered the market.
The AI drug development market had long been targeted as a future growth engine by U.S. big tech corporations such as Google and Nvidia, but as China jumps in, some say the competition for technological supremacy is growing fiercer.
◇ A "DeepSeek Moment" for bio AI as well
China's ByteDance Ltd. unveiled the open-source protein structure prediction model Protenix late last year and went on to announce Protenix v2 in Apr..
It is a model that goes beyond predicting the structure of the protein itself to more precisely forecast how a therapeutic antibody binds to a target protein—that is, the structure of the antibody–antigen complex. It has been advanced to more accurately predict interactions among diverse biomolecules, including not only proteins but also DNA, RNA, and small molecules.
In the industry, it has been called one of the most notable open-source models since AlphaFold in the field of biomolecule-based models, the core of AI drug development.
Seok Cha-ok, Galux CEO (a Seoul National University chemistry professor), said, "I was shocked by Protenix's performance," and added, "Just as DeepSeek appeared in the language model field, I felt a 'DeepSeek Moment' had arrived in bio AI." Seok noted, "I did not expect China to catch up this fast."
The DeepSeek Moment is a phrase likening the shock and paradigm shift when the Chinese AI startup DeepSeek built a high-performance AI model at low cost comparable to U.S. big tech. It has also been compared to the shock of the Soviet satellite launch in 1957, dubbed a "Sputnik moment" for AI.
In the bio industry, the emergence of Protenix is interpreted not as a simple AI model release but as a declaration of China's big tech entering bio.
◇ Beyond structure prediction, into the era of "drug design"
The opening of the AI drug development era began with AlphaFold, developed by Google DeepMind.
AlphaFold is an AI that predicts three-dimensional structures from only a protein's amino acid sequence. It was credited with effectively solving the long-standing life science challenge of protein folding and received the 2024 Nobel Prize in chemistry.
AlphaFold dramatically improved the accuracy of protein structure prediction and opened the AI drug development era. As Generative AI and protein language models advanced, AI has evolved beyond predicting protein structures to directly designing new small molecules, therapeutic antibodies, and proteins de novo.
Alphabet, Google's parent company, founded the AI drug development corporation Isomorphic Labs in 2021 based on this progress. The company has signed successive drug research deals with global pharmaceutical companies—Novartis and Eli Lilly and Company in 2024, and Johnson & Johnson (J&J) this year. In Feb., it unveiled its in-house AI drug design engine, IsoDDE, expanding its technology beyond protein structure prediction to directly designing candidates. It also recently raised $2.1 billion in investment.
CEO Seok said, "AI drug development is moving beyond simply predicting protein structures to the stage of designing new therapeutics," and explained, "If you build a single AI model that properly understands biomolecules, you can design everything from proteins and small molecules to cell therapies and gene therapies."
◇ Galux and Proteina accelerate commercialization of AI drug development
More corporations are targeting the AI drug development market. U.S. startup Chai Discovery, which collaborates with big pharma such as Novartis, Pfizer, and Eli Lilly and Company, and Nvidia, which launched the AI drug development platform BioNeMo, are among them.
Insilico Medicine, a company listed on the Hong Kong stock exchange, is also drawing attention in the pharmaceutical field. The company uses a Generative AI platform to conduct everything from target discovery to candidate design. It said AI has shortened the candidate discovery period, which used to take years, to around one year.
Korean corporations are also picking up speed.
Galux is expanding its business based on its AI protein design platform, GaluxDesign. After starting joint research on anticancer protein drugs with LG Chem(051910), the company has widened collaboration by signing successive deals for next-generation multi-specific antibodies with Celltrion(068270), an AI-based siRNA delivery platform with OliX Pharmaceuticals(226950), and joint development of a BBB (blood-brain barrier) protein delivery platform with Aimed Bio(0009K0).
Notably, after unveiling last year a world-first technology to precisely design novel antibodies at a commercial antibody level, it recently succeeded in novel antibody design for eight out of nine targets, securing performance among global frontrunners. It is conducting joint research with numerous corporations, including three global pharmaceutical companies, and was recently selected for a candidate discovery project by the Korea Drug Development Fund (KDDF), pushing to commercialize its in-house AI design pipeline.
Proteina recently began commercializing AI-based de novo antibody design technology together with a research team led by Baek Min-kyung, a professor in the Seoul National University School of Biological Sciences. AI designs new antibodies, and Proteina uses its ultrahigh-throughput protein analysis platform to rapidly verify whether they actually bind.
◇ AI drug race shifts from "design" to "validation"
If the early focus was on how accurately protein structures could be predicted, the center of gravity in AI drug development is now shifting to de novo antibody design—designing new antibodies from scratch with AI—and to technologies that raise design success rates.
CEO Seok said, "If the first competition was about 'whether you can design antibodies,' it is now moving to 'how high a success rate you can achieve,'" adding, "The next stage will be a race to design, with AI, GPCRs (cell membrane receptors) and other undruggable targets that are hard to tackle with existing drugs."
As AI performance converges upward, the battleground is seen not in design capability itself but in how quickly the thousands or tens of thousands of candidates designed by AI can be validated through experiments and connected to actual drugs.
Hwang Seong-taek, a Proteina director, said, "In the end, what matters is how quickly and cheaply you can experimentally validate the multitude of candidates generated by AI," noting, "Going forward, competitiveness will hinge more on high-speed, high-throughput validation platforms than on AI itself."
Experts say technologies that combine AI with automated experimental platforms—and further with quantum computing—will determine next-generation competitiveness in drug development.
Chung Jae-ho, head of the Yonsei Institute of Convergence Science and Technology, said, "The next stage after AI is quantum computing," adding, "Quantum computing is not a technology that replaces AI, but one that maximizes AI's computational capability."
Chung said, "The current limits of drug development match exactly the limits of computational capability," and predicted, "In 10 years, anticancer drugs designed by technology corporations like Nvidia, Google, and Microsoft could emerge, and a new industrial structure could take shape in which AI designs drugs and pharmaceutical companies manufacture them."