VUMC to develop AI technology for therapeutic antibody discovery

(Press-News.org) An ambitious project led by Vanderbilt University Medical Center investigators aims to use artificial intelligence technologies to generate antibody therapies against any antigen target of interest.

VUMC has been awarded up to $30 million from the Advanced Research Projects Agency for Health (ARPA-H) to build a massive antibody-antigen atlas, develop AI-based algorithms to engineer antigen-specific antibodies, and apply the AI technology to identify and develop potential therapeutic antibodies.

ARPA-H is an agency within the U.S. Department of Health and Human Services that supports transformative high-risk, high-reward research to drive biomedical and health breakthroughs to benefit everyone.

“Over the last few decades, monoclonal antibodies have started playing an important therapeutic role in a wide range of disease settings, but we’re just scratching the surface. Monoclonal antibody discovery has the potential to impact a lot of different diseases where currently there are no therapeutics,” said Ivelin Georgiev, PhD, professor of Pathology, Microbiology and Immunology, director of the Vanderbilt Center for Computational Microbiology and Immunology, and the project principal investigator.

Traditional methods for antibody discovery are limited by inefficiency, high costs and fail rates, logistical hurdles, long turnaround times and limited scalability, Georgiev said.

“What we’re proposing to do is going to address all of these big bottlenecks with the traditional antibody discovery process and make it a more democratized process — where you can figure out what your antigen target is and have a good chance of generating a monoclonal antibody therapeutic against that target in a very effective and efficient way,” said Georgiev, who is also professor of Biomedical Informatics, Computer Science, and Chemical and Biomolecular Engineering.

Antibodies are part of our immune system. They are proteins produced by white blood cells (B cells) that bind to and inactivate antigens — targets on viruses, bacteria and even our own cells. Antibodies are effective as preventive and therapeutic treatments against viruses, cancers, autoimmune disorders and other diseases.

To identify a candidate therapeutic antibody, researchers generally screen and test thousands of antibodies against an antigen target, looking for the “needle in the haystack” that binds to and neutralizes the target. The traditional discovery process requires specific types of biological samples. For example, to find antibodies against an infectious disease pathogen, blood samples from people or animal models exposed to the pathogen are required. And then, if the pathogen mutates, a therapeutic antibody may become ineffective.

“With a computational approach, you’re no longer dependent on access to biological samples or multiple screening cycles,” Georgiev said. “You can simulate variants and generate antibodies ahead of time before the variants arise.”

Georgiev and his colleagues are engaged in three tasks as they work toward developing computational approaches for antibody discovery:

Generation of an antibody-antigen atlas of unprecedented size and variety Development of AI-based algorithms for extracting information from the antibody-antigen atlas and engineering antigen-specific antibodies Proof-of-concept studies to apply the AI technology to identify antibody candidates against antigen targets of biomedical interest For the first task, the researchers are using a technology they developed called LIBRA seq (Linking B-cell Receptor to Antigen specificity through sequencing) that enables high-throughput mapping of antibody-antigen interactions for many antigens and B cells at the same time.

“For computational methods to work, we need to have a lot of data,” Georgiev said. “The scale of data that’s available for antibodies and antigens is lower than in other fields, which has been one of the limiting factors when it comes to developing AI approaches.

“If we train algorithms on the data that exists currently — much of it is for SARS-CoV-2, flu and HIV — the algorithms may be accurate for these targets, but they are less likely to be successful in extrapolating to a new target. We need to train them with a more diverse set of antigen targets, which is where LIBRA-seq comes into play.”

The investigators aim for the atlas to include hundreds of thousands — and potentially over 1 million — antibody-antigen pairs, compared to approximately 15,000 pairs currently available from published data, providing an unparalleled resource for researchers worldwide.

The team is already moving forward on the second task of building computational models, which they will improve as they populate the antibody-antigen atlas. For the third task, they will apply the AI technology to develop antibodies against cancer antigens and bacterial, viral and autoimmune targets. They will select one candidate antibody for preclinical development up to and including IND (investigational new drug) application.

“Our project will be providing a platform that can be used for a variety of different diseases, not just the specific targets we’re interested in,” Georgiev said. “Our team has spent many years trying to discover antibodies against a variety of indications, and it’s such an inefficient process with a lot of failure. If we can help change that, that’s going to be huge — not just for us, but for the entire field and for people with diseases where antibody therapies can make a difference.

“It’s going to be hard. It’s not an easy problem, but I think we have a good foundation for it, and we’ll do the best we can to make it work.”

Collaborators on the project are: Ben Ho Park, MD, PhD, Sarah Croessmann, PhD, Eric Skaar, PhD, MPH, Maria Hadjifrangiskou, PhD, and Jeremy Goettel, PhD, at VUMC; Tedd Ross, PhD, and Giuseppe Sautto, PhD, at Cleveland Clinic; and Maria del Pilar Quintana Varon, PhD, and Lars Hviid, PhD, at the University of Copenhagen. The Brock Family Center for Applied Innovation, a catalyst for advancing translational research to market, has engaged with and supported the Georgiev team.

Vanderbilt University and VUMC shared resources that are critical to the project are: VANTAGE (Vanderbilt Technologies for Advanced Genomics), ACCRE (Advanced Computing Center for Research and Education), and FCSR (Flow Cytometry Shared Resource). Wheeler Bio will participate in IND-enabling studies, cell lin

END