What if researchers could design antibodies with specific biological functions rather than simply search for molecules that bind a target? That question ran through The Antibody Series 2026 (TAS 2026), where speakers explored how computational design, AI, and increasingly complex antibody formats could expand what these therapies can do.
But the conference also underscored a persistent challenge. Mark van Dijk, Chief Scientific Officer at Fairjourney Biologics, told DDN that antibody technology may no longer be the main limitation. Instead, “the biology is the limitation.”
Researchers can increasingly design molecules with the stability, specificity, and potency they want, but knowing which biological interactions to target in the first place — and how to turn those interactions into an effective treatment — remains much harder.
Building antibodies by design
One emerging ambition is to design antibodies with desirable properties from the outset, rather than discovering candidates first and optimizing them later. The challenge is to improve stability and developability without losing the structural diversity needed to recognize a wide range of targets.
This is the thinking behind computationally designed antibody repertoires, an approach that Sarel Fleishmann, Professor at the Weizmann Institute of Science, discussed in his presentation at TAS 2026. In the past, synthetic repertoires have improved antibody discovery by providing a controlled, animal-free way to generate human antibodies. However, designing libraries around stable antibody frameworks can limit the structural diversity available for recognizing challenging targets, creating a trade-off between developability and discovery potential.
Fleishmann described CADAbRe, a structure- and energy-based approach that aims to address both challenges. The method uses hundreds of human antibody frameworks and designs compatible sequences for CDR H3, a region that plays an important role in recognizing antigens. The aim is to produce a library containing billions of potential antibodies that are both structurally varied and predicted to fold stably. According to Fleishmann's presentation, the team produced 2.5 billion antibody fragments on phage in 2026, with DNA synthesis costing €38,000.
Rather than simply expanding the number of antibodies that can be screened, CADAbRe is an attempt to make diversity more deliberate, incorporating structural constraints before the library reaches the experimental stage. The longer-term question is whether this kind of design can consistently produce useful binders across the targets that have proved difficult for conventional antibody discovery.
AI moves from structure to specificity
Computational tools in general were a major theme at the conference. Much of the recent progress in protein AI has centred on predicting what proteins look like or designing molecules with a particular structure. However, Sai Reddy, a Professor at ETH Zurich, focused on a different idea. Given the amino acid sequences of an antibody and an antigen, can a model predict whether the two will recognize each other?
This is known as the sequence-to-specificity problem, and it remains difficult to solve reliably at scale. Reddy told DDN that molecular specificity is a scarce asset in drug discovery. Finding an antibody with the right binding properties can take years of screening and experimentation, making specificity one of the key bottlenecks in developing new therapeutics. His goal is to use AI to make that search more efficient, allowing researchers to sift through large collections of antibodies and antigens and identify likely interactions without experimentally testing every possible pairing or first predicting their structures.
That is the idea behind CALM, or Cross-attention Adaptive Immune Receptor–Antigen Language Model. Rather than first predicting the structures of an antibody and antigen, the model learns from known antibody–antigen pairs to identify sequence patterns associated with recognition. It can then be used in both directions, asking which antigens an antibody might recognize or which antibodies might bind a particular antigen.
That could eventually open another route into antibody discovery. Instead of generating candidates and experimentally screening each one, researchers could use sequence-based models to narrow the search space first. For now, though, CALM represents an early step toward that goal rather than a replacement for experimental testing. Predicting molecular recognition from sequence alone remains a difficult problem, particularly when models encounter antibodies and antigens that differ substantially from those used for training.
Engineering biology, not just binding
Melissa Geddie, of Diagonal Therapeutics, spoke on harnessing multispecificity to drive novel functions, framing it as a way to move antibody engineering beyond simple binding and toward more deliberate biological design. She described how valency, multispecificity, and geometry can now be engineered by intention, giving rise to next-generation, multifunctional antibodies.
Antibody engineering, she noted, has evolved from monospecific binding to multispecific biology, with molecules designed to bring together elements such as two cells, bridging them, driving co-stimulation, or controlling precisely where activity takes place.
Geddie called this an exciting and vast era for multispecifics, one that serves less as an end in itself and more as an enabler to expand biology. She emphasized that true innovation isn't a matter of how many targets a molecule can hit, but of identifying the biological question researchers are actually asking, since that's what allows new biology to be unlocked in the first place. To illustrate this approach, she pointed to Radiant's Multabody platform, a multispecific, multi-affinity antibody format that goes beyond conventional bi- or trispecific designs. Rather than layering on additional targets, the modular platform is built around breakthrough antibody engineering aimed at generating genuinely new function.
Designing more complex biological interactions
The third major theme was the development of more complex antibody formats designed to produce biological effects that a conventional monoclonal antibody cannot achieve alone.
This was particularly evident in Joanne Hulme’s talk. Joanne, Chief Scientific Officer at Radiant Biotherapeutics, described the progression from monoclonal antibodies to bispecific and multispecific molecules, where researchers can combine multiple binding specificities within a single therapeutic. In these formats, binding is only part of the design equation. Researchers also need to consider how strongly and where a molecule binds, how multiple interactions work together, and how those interactions translate into a biological response.
Radiant’s Multabody platform combines multivalency and multispecificity within a single molecule. Rather than relying on one binding interaction, Multabodies can engage multiple epitopes on the same target or different disease-associated targets at the same time. Hulme said this can increase avidity, the overall strength produced by multiple simultaneous binding interactions, while allowing the molecule to engage more than one biological pathway.
Hulme highlighted the company’s lead program, RBT-101, which targets 4-1BB, a co-stimulatory receptor involved in T cell activation and persistence. Previous efforts to develop 4-1BB agonists, such as urelumab, have been limited by liver toxicity. In contrast, Radiant reported that RBT-101 produced complete responses in all treated animals in a colorectal cancer mouse model, with no signs of liver toxicity. Additionally, when the mice were rechallenged with fresh tumor cells three months after treatment, there was no detectable tumor growth, suggesting that the treatment had induced long-lived antitumor immunological memory.
Importantly, Hulme noted that stronger agonism does not necessarily translate into a better therapeutic effect. In the company’s experiments, urelumab produced a stronger 4-1BB agonist signal than RBT-101, yet did not produce complete responses in all animals or prevent tumor regrowth after rechallenge. The comparison highlights the importance of how a therapeutic engages its target, rather than simply how strongly it activates it.
For diseases such as cancer and infectious disease, where several pathways, cell types, or epitopes can contribute to disease, multispecific formats could provide researchers with more ways to control those interactions within a single therapeutic.
Delivering better therapies quicker
As a whole, TAS 2026 highlighted how antibody discovery is moving beyond the search for molecules that simply bind a target. Researchers are now developing new ways to design, predict, and engineer antibodies, with the aim of making drug discovery more efficient and bringing useful therapies to patients more quickly.
However, the growing ability to engineer increasingly sophisticated antibodies makes understanding the target itself even more important. A molecule can be designed to bind with high specificity or to engage several targets at once, but that does not necessarily tell researchers which interaction will produce the desired biological effect.
As Tariq Ghayur, Scientific Advisor for Fairjourney Biologics, told DDN during the conference, “Target biology is really key. A single target can have multiple functions, and antibodies are not natural ligands for the targets — they bind at different places. That's the work you have to do very early on: to understand which part of the target you should bind to, and with what properties.”
