TCRs targeting epitopes in infectious diseases or cancer play a central role in spontaneous and therapy-induced immune responses. Here, we built large phage display TCR libraries and screened them against the cancer testis antigen NY-ESO-1. We then trained a machine learning TCR-epitope interaction predictor on these data and identified several epitope-specific TCRs directly from TCR repertoires. Functional assays revealed activity towards the NY-ESO-1 epitope and no cross-reactivity with self-peptides.

PolyMap is a high-throughput method for mapping thousands of protein-protein interactions in a single tube. Here we probe antibody libraries isolated from human donors against a set of SARS-CoV-2 spike variants to demonstrate how PolyMap can be used to profile immune responses, map epitopes of hundreds of antibodies, and select functionally-distinct clones for therapeutics.

Nature has evolved millions of venom-derived peptides with diverse biological functions. Many of these peptides are stabilized by multiple disulfide bonds, giving them ideal pharmacological profiles. We have leveraged this natural diversity to create a new peptide discovery platform based on venom peptides using phage and yeast display and enhanced by machine learning technology. This venom-based platform offers significant advantages in functionality and developability compared to traditional peptide discovery methods, making it a powerful tool for discovering new peptide therapeutics.

The Bicycle platform uses proprietary bicyclic peptide phage display technology to deliver a unique toolkit of building blocks to create novel medicines. Bicycle molecules combine rapid extravasation and extensive tissue penetration with renal clearance and tuneable half-life. Bicycle peptides can target tumour antigens with different cytotoxic, radionuclide and imaging payloads. The Bicycle advantage provides opportunities to deliver tumour killing through different mechanisms, complemented by imaging to guide the therapeutic process.

In this presentation, I will provide insight into the newest developments within recombinant snakebite antivenoms and demonstrate how phage display technology can be used to find monoclonal antibodies and nanobodies with unprecedented efficacy in rodent model compared to the standard of care, namely antivenoms derived from the plasma of immunised animals. I will further provide perspectives for what is needed to advance recombinant antivenoms into the clinic.

In recent years, the convergence of Cryo-Electron Microscopy (Cryo-EM) with Artificial Intelligence (AI) has revolutionised the landscape of structural biology and antibody engineering. This talk explores the innovative use of Cryo-EM for biologics enhanced by AI and Physics simulations to accelerate the antibody lead optimisation process. By capturing high-resolution structural data on antibody-antigen complexes, Cryo-EM provides unparalleled insights into binding interactions and mechanisms of action. Coupling this with AI algorithms capable of rapid identification of mutations that enhance binding affinity, specificity, and stability. This presentation will highlight case studies where this synergistic approach has led to breakthroughs in optimising antibodies for complex targets, including those considered challenging due to epitope variability or conformational flexibility. With Cryo-EM's ability to handle dynamic molecular ensembles and AI's power in data-rich environments, the integration of these technologies promises unprecedented advancements in therapeutic antibody development.

Therapeutic antibodies require structural optimisation, often guided by cryo-EM data. We present CrAI, the first fully-automated method to detect antibodies in cryo-EM maps using machine learning and a custom database. CrAI identifies Fab and VHH fragments in seconds, even at resolutions up to 10 Å, without additional inputs. It significantly outperforms existing tools in speed and accuracy, enabling seamless integration into structural analysis pipelines.

We are developing two complementary methods combining atomistic design and machine learning for one-shot antibody optimisation. Our work demonstrates that designing the antibody variable fragment for stability can improve antibody affinity and developability without prior experimental data and with twenty or fewer designs tested. We can additionally generate small libraries containing thousands of low-energy designs that can be screened for further improvements and used as training data for machine learning. Finally, I will describe our work developing an approach for atomistic design of synthetic antibody repertoires based on hundreds of antibody frameworks for accelerated discovery of developable antibodies.

In the past years we have been focusing on development of antibodies leads using phage display in complex selections. Many therapeutic targets are normally found in membranes; in order to ensure binding of antibodies to native, therapeutically-relevant epitopes, selection for binding specific targets is best performed when the antigens are presented in their native environment.