Paving the road to functionally active antibody drugs

Antibody drugs have reshaped how we treat diseases. By harnessing the immune system to destroy tumours or temper excessive inflammation in autoimmune disorders, these therapeutics have become indispensable tools in modern medicine. Their unique properties, such as high specificity and the ability to engage the immune system, have enabled breakthroughs in previously untreatable conditions.
Despite their success, most antibody drugs are limited to inactivation, target inhibition, or recruiting immune functions to specific sites. This leaves much of their potential untapped. Functionally active antibodies that activate or modulate the activity of their targets represent a new frontier. These antibodies can address unmet needs by accessing novel therapeutic mechanisms, but are notoriously difficult to discover. Unlocking these transformative drugs requires moving beyond conventional discovery to leveraging high-throughput functional assay technologies for generating the large datasets required for machine learning (ML).
The promise of antibody drugs
Antibodies are essential components of the immune system, evolved to neutralise pathogens, toxins, and foreign molecules. Their variable regions bind targets with precision, while constant regions recruit immune responses to eliminate threats.
Antibody drugs exploit these binding and immune-related functions to combat a wide range of diseases. Starting with the first monoclonal antibody, muromonab-CD3, used for organ transplant rejection, these therapies have expanded into oncology, autoimmune diseases, and infectious diseases. In cancer, antibodies can target tumour cells directly for destruction (e.g., antibody-drug conjugates) or amplify the immune response (e.g., ADCC, CDC, bispecifics). In autoimmune diseases, they selectively block pro-inflammatory signals or receptor activity, halting disease progression without suppressing the immune system broadly. Recently, antibodies have played a pivotal role in combating viral infections, including COVID-19, where their precision and scalability have proven vital.
Antibodies as a drug modality also offer distinct advantages over traditional small molecules or peptides. Their specificity minimises off-target effects, improving safety profiles. Their large size and structural complexity allow them to interact with diverse protein surfaces, including complex targets inaccessible to smaller drugs. Additionally, antibodies are safely metabolised and have long half-lives, reducing dosing frequency and making them ideal for chronic conditions.
Current methods and challenges in antibody discovery
Antibody discovery involves searching for sequences with desired traits, such as high affinity or biological activity. Traditional methods like hybridoma technologies, which isolate antibodies produced by immunised mice, remain foundational. Here, the mammalian immune system naturally refines antibodies for specificity and affinity through processes like somatic hypermutation. This approach has been complemented by phage and yeast display technologies, which allow in vitro screening of large libraries to identify binders. Recent advancements have expanded these methods, such as incorporating transgenic animals that produce human antibody repertoires and engineering synthetic libraries with optimised diversity. These innovations increase the pool of potential candidates and improve the likelihood of discovering hits.
However, these methods are largely geared toward finding binders with high affinity, not potent function. This limitation stems from their natural function: antibodies evolved to neutralise pathogens, not induce structural changes in therapeutic targets. Functional antibodies capable of modulating signalling or activating pathways require an entirely different discovery approach.
To identify these rare functional antibodies, researchers rely on functional screening methods, such as cell-based assays or in vitro reconstituted systems that mimic target pathways. These assays measure activities like receptor activation or inhibition, providing insights into the biological effects of antibody binding. Scaling these methods is critical to finding rare hits.
Technologies like microfluidics and droplet-based systems have further increased throughput, enabling researchers to screen hundreds of thousands to millions of variants for function. Synthetic biology has pushed the boundaries even further, with engineered cells capable of reporting the functional activity of hundreds of millions of antibody variants.
Why scale matters
Functionally active antibodies are exceptionally rare. Out of the vast pool of potential antibody sequences, only a tiny fraction will interact with targets, and then only a subset of that fraction will interact with the target in ways that induce structural changes and alter target activity. For example, to activate G-protein-coupled receptors (GPCRs), a common but challenging target class for antibodies, antibody binding needs to mimic natural ligand binding, or bind multiple allosteric sites on the target in a precise relative orientation to promote active conformations on the intracellular side. Such antibodies are exceedingly difficult to identify, as they typically do not have the highest affinity interaction and thus are missed by affinity-based discovery methods. High-throughput activity measurement is essential to uncover these elusive candidates.
Encouragingly, progress has been made with more accessible target classes, such as receptor tyrosine kinases and cytokine receptors. These proteins are activated by dimerisation or clustering, rather than precise structural modification. They have large extracellular domains, making it easier to find antibodies that bind them; fragments of these binding antibodies can be combined, such as in a bispecific antibody format, to cluster these receptors upon binding and activate them much like the natural ligands.
The power of AI in antibody discovery
The intersection of data and computation is changing antibody discovery. AI algorithms can process vast datasets, uncover patterns, and predict outcomes, making them ideal for exploring the vast combinatorial space of antibody sequences. By integrating data on binding, structure, and activity, AI can usher the transition from using data for discovery, a fundamentally probabilistic process, to using data to learn how to design, an intentional engineering process. In the late 90s, protein structure prediction was one of the last great unsolved biology problems. Breakthroughs like AlphaFold have showcased AI’s potential to predict protein structures with unprecedented accuracy. AI is now being used to produce “de novo” antibodies that can bind desired targets. Even if only a fraction of the generated antibody binders validate experimentally, the fact that a reasonable fraction do validate is an astounding feat.
However, prediction of active antibody sequences remains an open challenge. While mapping sequence to structure and structure to interactions is now becoming feasible, predicting sequences that translate into functional outcomes is far more complex. Activity depends on dynamic protein structures and interactions between them, and these interactions are influenced by myriad biological factors.
High-quality activity data is the critical bottleneck. AlphaFold works because it trained on huge public datasets like the Protein Data Bank (PDB). Natural language processing and image recognition were also “solved” by AI with large amounts of text and image data now publicly available. Analogously, antibody ML models require large, high-quality training datasets linking sequence, structure, and activity. Most activity assays generate orders of magnitude less than the tens of millions of parameters in protein ML models. Only recently have cell-based technologies begun to generate hundreds of millions of activity datapoints sufficient to train predictive models. Data, not computational power, is the key to unlocking the next generation of activating and modulating antibodies.
The road ahead
While challenges remain, the potential for functionally active antibodies is immense. These drugs offer solutions to previously untreatable conditions by modulating biological pathways inaccessible to other drug modalities. By combining traditional antibody engineering, synthetic biology, and advanced computational modelling, researchers are building technologies to find antibodies capable of addressing the most challenging biological targets and complex diseases.