In many fields of endeavor, traditional techniques are being turbocharged by rapid, rational, computer-driven techniques—that is, by artificial intelligence (AI). And there is plenty of evidence that one of these fields is monoclonal antibody (mAb) discovery. It has long benefited from applications of wet chemistry and analytics. And now it is starting to benefit from AI, too. For example, AI technologies are already looking for novel antibody sequences that can bind to specific targets. In addition, AI technologies promise to address many development-stage issues such as manufacturability.
Without AI, antibody discovery can be a slog. “The problem with how antibodies are discovered—and how they are optimized for high affinity, high stability, low viscosity, and long in vivo half-life—is that there is too much trial-and-error testing,” says Peter M. Tessier, PhD, a professor of chemical biology at the University of Michigan.
AI reduces the need for experimentation by identifying high-quality mAb candidates much more rapidly, at the discovery stage. Through its potential to optimize several properties simultaneously, AI could help narrow the number of mutations required to improve an antibody’s performance to render it ready for clinical testing.
Today, the benefits of AI-based mAb discovery, Tessier says, have less to do with de novo design of new antibodies, and more to do with “the rapid, efficient identification of antibodies with impressive combinations of properties in a much shorter timeframe.”
Finding new chemistry
Many experts believe that AI-based mAb design will enable developers to exploit the structures and properties of diverse antibody sequences, and to propose novel chemistry design space, and possibly to address unmet medical needs. One such expert is Peyton Greenside, co-founder and CSO, BigHat Biosciences.
“Traditional antibody engineering is performed manually—that is, designed by a scientist—or through biased- or random-sequence mutagenesis—for example, through an affinity maturation library,” says Greenside. Developers can only expect to improve or fix a relatively small number of antibody properties or deficiencies, for example, affinity, purity, or viscosity.
In this zero-sum game, gains in one property usually come at the expense of others, and improvements are restricted to the sequence space within a few mutations of the starter sequence. Plus, standard approaches have difficulty conceiving of novel chemical spaces or formats other than standard mAbs.
“AI-based approaches can work with any number of existing sequences, model many properties simultaneously, generalize to more diverse sequence space, and generate or propose novel sequences predicted to meet certain design conditions,” Greenside continues. “These algorithms can ingest more data and model more simultaneous complexity than a typical scientist can hold in their brain at once.”
Based on their ability to learn underlying sequence and/or structural determinants of antibody properties across thousands to millions of sequences, AI models can more adeptly propose potentially more diverse solutions to solve a challenging engineering problem or, in the case of de novo discovery, to design an antibody from scratch.
What it takes
“Machine learning–guided protein design is exciting and growing, with new capabilities emerging monthly,” BigHat’s Greenside points out, “but ultimately the proof of these techniques will be their ability to get antibodies into the clinic to treat patients. As we develop more and more sophisticated approaches, the most exciting applications are yet to come.”
It may seem remarkable to old-school chemists and biologists, but the only requirements for designing antibody sequences and structures today are a laptop, open-source software, and software libraries. The emergence of AI provides an added dimension, mainly the ability to conceive and deliver instructions for previously unknown antibody structures and functions, that is, to probe unmined chemical space.
But while AI-based antibody discovery involves a relatively modest investment in hardware and software, an exceptionally high level of skill is required to program and maintain the software, to design experiments, and to decide which molecules to promote.
“At a minimum, one would need skills with Python, or in a deep learning framework like PyTorch, plus facility with and Biopytthon and ANARCI to deal with antibody sequences,” Greenside observes. “Also, to overcome data scarcity in de novo antibody design, implementers of AI must tap into general antibody or structure predictors like AlphaFold and IgFold, protein language models like ESM or AntiBERTy, a diffusion model for protein design like RFdiffusion, or even rational design tools like Rosetta and the structure-viewing program PyMol. All of this can be run comfortably on a single GPU on a decent laptop.”
However, since models cannot perfectly predict how AI-designed antibodies perform in real life, a wet lab is essential to validate any computational designs and to test for stability, aggregation propensity, target affinity, purity, and other properties.
Coalescing competencies
So, rather than eliminate or reduce reliance on wet chemical assays and attendant analytics, AI-based mAb ups the requirements of these tools for speed, robustness, and reliability. Earlier this year, ImmunoPrecise Antibodies (IpA), a Canadian AI-biopharm research and technology company, made a point of not only purchasing Carterra’s surface plasmon resonance (SPR) instrument platform, but also formally announcing the acquisition as part of its commitment to AI. SPR, a rapid biophysical assay for quantifying molecular affinity, provides affinity data in seconds and is highly sparing of test materials.
At the time, IpA’s CEO Jennifer Bath, PhD, noted that SPR would complement her company’s “high-throughput antibody discovery, production, and screening capabilities” and get the company “closer to making the fastest and most cost-effective drug discovery workflow.”
Not only are there AI-oriented antibody discovery companies acquiring analytical and wet lab competencies, but there are also traditional antibody discovery organizations acquiring AI competencies. Consequently, GEN expects more formal relationships to develop between AI firms and companies with established discovery and development expertise.
For example, in April 2024, BigHat Biosciences entered a collaboration with Johnson & Johnson subsidiary Janssen Biotech to combine the larger firm’s drug discovery, clinical development, and data science expertise with BigHat’s Milliner platform, a suite of machine learning technologies integrated with a high-speed wet lab, to guide the design and selection for high-quality antibodies for multiple neuroscience therapeutic targets.
Milliner integrates a synthetic biology–based high-speed wet lab with machine learning technologies into a complete antibody discovery and engineering platform, with the goal of engineering antibodies with more complex functions and improved biophysical properties. This approach, according to BigHat, reduces the difficulty of designing therapeutic proteins to treat a range of chronic and life-threatening illnesses while massively speeding up candidate discovery and validation.
While more relationships between traditional biopharma and AI are in the works, joint ventures will moreover not be limited to traditional biopharma and AI companies, as diverse service companies will also seek to combine forces to provide AI-enhanced design and discovery services.
For example, in March 2024 BioGeometry, a digital biology firm specializing in AI-driven protein design, and Sanyou Biopharmaceuticals, a biologics R&D services company, joined forces to leverage their respective expertise to create a next-generation antibody drug discovery platform. Under this agreement, Sanyou will integrate its wet chemistry expertise with GeoBiologics, BioGeometry’s generative AI antibody design platform.
In late 2023, Absci, an AI antibody discovery company, partnered with AstraZeneca to employ AI in the search for new cancer treatments. The agreement capitalizes on Absci’s AI-based protein design platform with AstraZeneca’s oncology expertise to expedite discovery and development.
Absci’s Integrated Drug Creation platform combines generative AI and scalable wet-lab technologies based on millions of protein-protein interactions. AI is employed at the front end (discovery) with more-or-less conventional analytics and chemical assays, and it has the potential to channel discovered proteins to the clinic more quickly. A six-week-long workflow is anticipated.
The announcement follows Absci’s publication on the design and validation of de novo antibodies using the company’s “zero shot” generative AI model. With this model, developers can design antibodies to bind to specific targets without the need for training data from antibodies known to bind to those targets. In other words, the target is the only requirement. This allows the generation of antibodies structurally and chemically unlike those in existing antibody databases, including versions of all three heavy-chain complementarity-determining regions most critical to target binding. Absci claims that its discovery platform finds up to 30 times as many hits as those found with standard antibody library methods.
Improving all characteristics
In antibody discovery, AI has focused on the identification of molecules with high target affinity and novel chemical space. However, AI also has the potential to improve almost any antibody attribute. One such characteristic is developability—the likelihood that a molecule will progress smoothly through the chemistry, manufacturing, and control (CMC) process at reasonable cost and within reasonable timelines.
In a paper published in early 2023, scientists at Shanghai-based WuXi Biologicals proposed AI-based screening to “lower the risk that an antibody candidate with poor developability will move to the CMC stage,” and that this should be done “as early as possible … in a rapid and high-throughput manner while consuming small amounts of testing materials.” This approach, the scientists argued, can be applied not just to mAbs but to advanced therapies like bispecific antibodies, multispecific antibodies, antibody-drug conjugates, and other mAb derivatives.
