AI Breakthrough: Scientists Design Functional Antibodies from Scratch Using Machine Learning

AI-designed antibodies created from scratch

Research led by the University of Washington reports on an AI-guided method that designs epitope-specific antibodies and confirms atomically precise binding using high-resolution molecular imaging, then strengthens those designs so the antibodies latch on much more tightly. Why epitope targeting matters Antibodies dominate modern therapeutics, with more than 160 products on the market and a projected value of US$445 billion in 5 years. Antibodies protect the body by locking onto a precise spot -- an epitope -- on a virus or toxin. That pinpoint connection determines whether an antibody blocks infection, marks a pathogen for removal, or neutralizes a harmful protein. When a drug antibody misses its intended epitope, treatment can lose power or trigger side effects by binding the wrong target. In medical development, knowing exactly where an antibody lands on a molecule can decide whether it succeeds in patients or fails in trials. Researchers design epitope-specific antibodies to target disease-critical regions, such as the receptor-binding tip of a virus spike or the toxic domain of a bacterial protein. Reaching that level of accuracy normally requires a slow, iterative process with years of lab work, involving animal immunization, multiple rounds of screening, and structural studies to confirm the binding site. A reliable way to plan those interactions on a computer could make antibody creation faster and more focused, aiming at the precise molecular surfaces that control infection, toxicity, or cell signaling. Computational efforts have largely optimized existing antibodies, and some deep learning models have proposed variants once a binder already exists. Recent generative approaches have needed a starting binder, leaving de novo, epitope-specific antibody creation as an unmet goal. In the study, "Atomically accurate de novo design of antibodies with RFdiffusion," published in Nature, researchers trained an AI system to build antibodies that recognize user-specified molecular sites. The model, known as RFdiffusion, used information about antibody frameworks and target surface "hotspots" to shape new binding loops. A second network, RoseTTAFold2, predicted whether each design would fold and bind as intended, filtering out unstable or misaligned candidates. Llama assisted research Design efforts focused first on single-domain antibodies known as VHHs. These miniature antibodies come from animals such as llamas and alpacas and are prized in research for being stable, compact, and easy to engineer. Their small size allows them to reach crevices on viral or bacterial proteins that full-size antibodies cannot. Researchers used a humanized VHH framework as the scaffold and aimed designs at influenza hemagglutinin, Clostridium difficile toxin B, RSV sites I and III, and the SARS-CoV-2 receptor-binding domain. Laboratory screens used yeast surface display to test 9,000 designs per target and E. coli expression with single-concentration surface plasmon resonance to evaluate 95 designs per target. Each target posed a distinct structural challenge, from the smooth surface of influenza to the complex folds of C. difficile toxin. Hitting the targets Influenza designs produced several lab-made antibodies that attached to the virus protein in test tubes. High-resolution imaging showed one of those matches lining up with the computer's plan at near-atomic detail, including how a key loop on the antibody reached the target site. Microscopic sugar on the virus shifted aside when the antibody settled in, a movement seen in the images and consistent with the planned approach. C. difficile toxin work yielded a compact antibody that grabbed the intended site and blocked a previously designed competitor from landing there. Lab tests on cells showed protection against the toxin's damage. Follow-up imaging captured the same docking behavior before and after lab evolution, indicating that improvements in grip did not change where or how the antibody latched on. Missing a few SARS-CoV-2 tests produced a compact antibody that attached only when the virus protein moved into the "up" position and blocked a known competitor at that spot. Imaging placed the connection on the right site while revealing a different angle of approach than planned, a result labeled a design failure by the authors. Designs aimed at a cancer-related peptide on human immune proteins showed on-target recognition in two separate assays, yet engineered T cells built from those designs did not kill tumor cell lines in lab tests. A good start Reported success rates remain low at 0% to 2% across targets, and authors point to improved filtering with AlphaFold3 ipTM as a route to enrichment. Prospects include faster and potentially more targeted antibody discovery as models and filters improve, with particular value for applications needing precise epitope engagement such as receptor-ligand blockade, conformational modulation, and conserved viral sites.

AI-designed antibodies promise big boost to drug development

A team led by Nobel Prize-winning scientist David Baker has used an artificial intelligence tool to create new functional antibodies in a breakthrough that could speed up drug development using the cutting-edge technology. In a study published on Wednesday in Nature, researchers at the University of Washington showed how a generative AI model could be used to design entirely new antibodies -- proteins that our immune systems produce to block infections -- from scratch. The pharmaceutical industry widely uses antibodies to create drugs, such as for cancer and coronaviruses. But creating the proteins using computers has been a long-standing challenge in drug design. Baker, who was a co-winner of the Nobel Prize in Chemistry last year for his work on computational protein design, called the breakthrough a "step change" for the pharmaceutical industry. "This is part of a big transformation in bioengineering of going from random library selection methods and evolution methods to rational design," Baker said. Conventionally, discovering antibodies is expensive, laborious and time-consuming, requiring the use of animal immunisation tests and extensive screening. Scientists wait months for the test animal to produce an antibody response, and the work involves lots of trial and error. Using an AI tool could speed up this process to weeks, without the need for animal testing, and could help scientists find more precise targets. The researcher's AI model -- dubbed RFantibody -- is based on a previous model by the same lab that designs new proteins, and is fine-tuned with additional data on antibodies. It was trained to generate new antibodies and predict which ones should be tested further in a lab, said Joe Watson, a researcher at the University of Washington and co-founder of Xaira Therapeutics. In the study, the team were able to design antibodies that successfully bound to an actual cancer protein. Finding antibodies for cancer is hard because often the difference between a tumour cell and a normal one might be a single protein. "Because we know where a piece is, we can just click on that, and tell the model, 'this is where I want an antibody that binds,'" said Watson. "That's why we think this is the future." The creation of functional antibodies using AI was a "remarkable achievement", said Francesco Aprile, associate professor in biological chemistry at Imperial College London. "Antibodies are important molecules in research and medicine, but developing them is often slow and difficult," he said. "This is an important step forward that could make antibody development faster and help drive progress in biotechnology and healthcare." The AI-boosted approach will now need to be tested on whether it can produce viable finished treatments for the diseases tested and other targets. The faster design of antibodies is just one part of the lengthy process of drug development that can take many years. It will not speed up time-consuming later elements, notably clinical trials and regulatory approvals.

AI can speed antibody design to thwart novel viruses

Artificial intelligence (AI) and "protein language" models can speed the design of monoclonal antibodies that prevent or reduce the severity of potentially life-threatening viral infections, according to a multi-institutional study led by researchers at Vanderbilt University Medical Center. While their report, published in the journal Cell, focuses on development of antibody therapeutics against existing and emerging viral threats, including RSV (respiratory syncytial virus) and avian influenza viruses, the implications of the research are much broader, said the paper's corresponding author, Ivelin Georgiev, Ph.D. "This study is an important early milestone toward our ultimate goal -- using computers to efficiently and effectively design novel biologics from scratch and translate them into the clinic," said Georgiev, professor of Pathology, Microbiology and Immunology, and director of the Vanderbilt Program in Computational Microbiology and Immunology. "Such approaches will have a significant positive impact on public health and can be applied to a broad range of diseases, including cancer, autoimmunity, neurological diseases, and many others," he said. Georgiev is a leader in the use of computational approaches to advance disease treatment and prevention. Perry Wasdin, Ph.D., a data scientist in the Georgiev lab, was involved in all aspects of the study and is the first author of the paper. The research team, which included scientists from around the country, Australia and Sweden, showed that a protein language model could design functional human antibodies that recognized the unique antigen sequences (surface proteins) of specific viruses, without requiring part of the antibody sequence as a starting template. Protein language models are a type of large language model (LLM), which is trained on huge amounts of text to enable language processing and generation. LLMs provide the core capabilities of chatbots such as ChatGPT. By training their protein language model MAGE (Monoclonal Antibody Generator) on previously characterized antibodies against a known strain of the H5N1 influenza (bird flu) virus, the researchers were able to generate antibodies against a related, but unseen, influenza strain. These findings suggest that MAGE "could be used to generate antibodies against an emerging health threat more rapidly than traditional antibody discovery methods," which require blood samples from infected individuals or antigen protein from the novel virus, the researchers concluded. Other Vanderbilt co-authors were Alexis Janke, Ph.D., Toma Marinov, Ph.D., Gwen Jordaan, Olivia Powers, Matthew Vukovich, Ph.D., Clinton Holt, Ph.D., and Alexandra Abu-Shmais.

Nobel winner's lab notches another breakthrough: AI-designed antibodies that hit their targets

Researchers from Nobel Laureate David Baker's lab and the University of Washington's Institute for Protein Design (IPD) have used artificial intelligence to design antibodies from scratch -- notching another game-changing breakthrough for the scientists and their field of research. "It was really a grand challenge -- a pipe dream," said Andrew Borst, head of electron microscopy R&D at IPD. Now that they've hit the milestone of engineering antibodies that successfully bind to their targets, the research "can go on and it can grow to heights that you can't imagine right now." Borst and his colleagues are publishing their work in the peer-reviewed journal Nature. Before the advent of AI-based tools, scientists made antibodies by immunizing animals and hoping they would produce useful molecules. The process was laborious and expensive, but tremendously important. Many powerful new drugs for treating cancer and autoimmune diseases are antibody-based, using the proteins to hit specific targets. Baker, who won the Nobel Prize in Chemistry last year, was recognized for his work unraveling the molecular design of proteins and developing AI-powered tools to rapidly build and test new ones. The technology learns from existing proteins and how they function, then creates designs to solve specific challenges. In the new research, the team focused on the six loops of protein on the antibody's arms that serves as fingers that grab its target. Earlier efforts would tweak maybe one of the loops, but the latest technology allows for a much bigger play. "We are starting totally from scratch -- from the loop perspective -- so we're designing all six," said Robert Ragotte, a postdoctoral researcher at IPD. "But the rest of the antibody, what's called the framework, that is actually staying the same." The hope is that by retaining the familiar humanness of most of the antibody, a patient's immune system would ignore the drug rather than mount an offense against an otherwise foreign molecule. The researchers tested their computer creations against multiple real-world targets including hemagglutinin, a protein on flu viruses that allow them to infect host cells; a potent toxin produced by the C. difficile bacteria; and others. The lab tests showed that in most cases, the new antibodies bound to their targets as the online simulations predicted they would. "They were binding in the right way with the right shape against the right target at the spot of interest that would potentially be useful for some sort of therapeutic effect," Borst said. "This was a really incredible result to see." Borst added that the computational and wet lab biologists worked closely together, allowing the scientists to refine their digital designs based on what the real-life experiments revealed. The software used to create the antibodies is freely available on GitHub for anyone to use. Xaira Therapeutics, a well-funded biotech startup led by IPD alumni, has licensed some of the technology for its commercial operations and multiple authors on the Nature paper are currently employed by the company. While the antibodies created as part of the research demonstrated the software's potential, there are many more steps to engineering a potential therapy. Candidate drugs need to be optimized for additional features such as high solubility, a strong affinity for a target and minimizing immunogenicity -- which is an unwanted immune response. Before joining IPD four years ago, Ragotte was a graduate student doing conventional antibody discovery and characterization using animals. The idea that one day you could get on a computer, choose a target, and create a DNA blueprint for building a protein was almost unimaginable, he said. "We would talk about it, but it didn't even seem like a tractable problem at that point." The Nature study is titled "Atomically accurate de novo design of antibodies with RFdiffusion." The lead authors include Nathaniel Bennett, Joseph Watson, Robert Ragotte, Andrew Borst, DéJenaé See, Connor Weidle and Riti Biswas, all of whom were affiliated with the UW at the time the research was conducted, and Yutong Yu of the University of California, Irvine. David Baker is the senior author. Additional authors are: Ellen Shrock, Russell Ault, Philip Leung, Buwei Huang, Inna Goreshnik, John Tam, Kenneth Carr, Benedikt Singer, Cameron Criswell, Basile Wicky, Dionne Vafeados, Mariana Sanchez, Ho Kim, Susana Torres, Sidney Chan, Shirley Sun, Timothy Spear, Yi Sun, Keelan O'Reilly, John Maris, Nikolaos Sgourakis, Roman Melnyk and Chang Liu.