Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • ADVERTISEMENT FEATURE Advertiser retains sole responsibility for the content of this article

Intelligent design of next-generation antibodies

Antibodies are a wonderful example of what natural selection, operating over millions of years of evolution, can engineer. Drug developers searching for novel, non-natural antibody therapeutics, however, typically need shorter lead times for their projects and, more importantly, increased probability of success. Nabla Bio is developing an innovative integrated artificial intelligence (AI) and wet-lab platform to expedite the design of new antibodies that not only bind their desired target with the requisite affinity, but which also possess the ‘developability’ properties that are crucial for the clinical and commercial success of antibody drugs.

Historically, developers of antibody drugs have tended to focus on the affinity of new antibodies for their desired target, while the issue of developability properties has been left as an afterthought—one often prompted by problems encountered in the path from the lab to the clinic. Although there has been considerable progress overcoming some of these challenges for monospecific antibodies, newer antibodies incorporating bispecific and multispecific designs pose a much harder problem as they typically have less-favourable biophysical characteristics. Nabla Bio is working on a solution to this development challenge, building on its co-founder Surge Biswas’s graduate work in the lab of George Church at Harvard Medical Schoolin Boston.

Across the antibody-drug industry, there is a recognized need for dedicated platforms that can identify candidate antibodies with both good affinity and developability properties early in the discovery process, before time and money have been invested in a candidate that is destined to fail down the line.

Dual-armed platform

Nabla Bio is meeting this need with a platform comprising two arms. The first is an innovative AI for designing antibodies of all formats (monospecific, bispecific and multispecific) that is differentiated from other AI-guided approaches in several ways. To start, using protein language and structure modeling technologies pioneered by Nabla scientists1,2,3, Nabla’s AI is trained on billions of natural protein sequences, including a few hundred million antibody sequences and several thousand antibody structures. This AI can then be used to incorporate experimental data and perform data-driven optimization of multiple antibody properties. Remarkably, for some universal protein properties such as stability, initial experimental data are often not required for engineering—an approach called zero-shot design. Finally, Nabla’s AI comprises generative models, which are used to generate libraries of antibody sequences that are enriched for good developability, and leverage antigen structure to design target- and even epitope-specific binders (Fig. 1).

Nabla’s antibody design process

Fig. 1 | Nabla’s antibody design process.

These libraries, ranging in size from 10,000 to 1,000,000 antibody designs, are then assayed in multiplex using a proprietary display system, with the coding sequence of each antibody carrying a unique DNA barcode tag so the performance of each design can be easily tracked with next-generation sequencing. In this stage, the affinity of each antibody to various forms of the target—human, mouse or monkey, for example—can be quickly measured in parallel, and at different concentrations, to create a whole suite of binding curves in one go, rather than compiling them one at a time.

A major difference between Nabla’s testing arm and those already in use is that it not only measures affinity, but also a range of properties that affect developability, including stability, polyspecificity, and expressibility. The performance of each AI-generated, DNA-tagged antibody on these various measures is then fed back into the AI model, which then generates another round of improved designs for further testing. Nabla then uses its AI to comb through this multiproperty dataset to select a small set of a few dozen lead molecules for characterization using industry-standard assays for binding, developability, and immunogenicity. Once a set of candidates meets recognized assay thresholds, engineering is considered to be over and the molecules are ready for preclinical and clinical testing.

Nabla’s AI capabilities have been extensively validated and formed the basis for a number of collaborations with pharma. Meanwhile, strong proof-of-concept data on the wet-lab components have confirmed the promise of this platform. With these integrated AI and wet-lab arms, Nabla can identify high-affinity antibodies with the developability characteristics required for clinical and commercial success. Nabla is currently establishing collaborations with select partners to co-develop next-generation antibody therapeutics, and welcomes discussions with potential future partners working in this space.


  1. Alley, E. C. et al. Nature Methods 16, 1315–1322 (2019).

    Article  PubMed  Google Scholar 

  2. Biswas, S. et al. Nature Methods 18, 389–396 (2021).

    Article  PubMed  Google Scholar 

  3. Chowdhury, P. et al. BioRxiv

Download references


Quick links