How we helped a leading pharma company accelerate nanobody discovery with GenAI

Challenge: Increasing the success rate of nanobody design

Nanobodies, small, single-domain antibody fragments, are emerging as a promising class of therapeutics thanks to their stability, manufacturability, and ability to reach targets inaccessible to conventional antibodies. However, designing them is still highly complex. The sequence space is vast, relationships between amino acids and biological properties are not yet fully understood, and discovery pipelines rely heavily on costly, trial-and-error lab screening.

Our client, a leading pharmaceutical company specializing in protein therapeutics, wanted to accelerate the design of nanobody candidates, increase candidate quality, and reduce the experimental burden.

How we helped

In close collaboration with the client's R&D experts, we designed a Lab-in-the-Loop Artificial Intelligence framework tailored to the scientific complexity of nanobody design and the practical realities of experimental workflows. Together, we developed a generative AI model tailored to tackle one of the foundational challenges in protein design: generating protein sequences that are expressible in vitro. The model is specifically trained to propose nanobody sequences with a high likelihood of successful expression, significantly improving the hit rate of viable constructs and reducing the burden on wet-lab screening.

The impact: Unlocking faster, smarter nanobody discovery with AI

With nanobodies gaining momentum as a next-generation therapeutic modality,  this collaboration marks a significant step toward positioning our client as a leader in nanobody design.

Near 100% expression rate of generated candidates

In the very first round of sequence generation and lab validation, the model proposed nanobody sequences with an expression success rate above 80%. By further incorporating existing experimental data into the model, the system achieved a near 100% expression rate.

Accelerated discovery of high-affinity candidates

A considerable share of the generated candidates was experimentally confirmed as functional binders, with a few of them even exhibiting nanomolar affinity to the target. This demonstrates the potential of data-driven design to significantly accelerate the development of therapeutic proteins.

Reduced burden and cost of experimental screening

Because fewer, higher-quality sequences need to be experimentally validated, the approach reduces experimental burden and frees up resources for more sensitive and resource-intensive assays that are currently impractical at scale. This enables the deeper validation of promising molecules, improving the overall efficiency and precision of the discovery process.

‍