Throughout 2022, Exscientia has progressed AI-driven capabilities for virtual biologics design, such as human antibodies, and is now establishing an automated biologics laboratory in Oxford to internally generate and profile novel antibodies.
Current approaches to optimise antibodies, even those that use machine learning, still depend on discovering antibodies by experimental screening methods. Combining generative AI design and virtual screening of biologics would allow investigation of a broader antibody space and support Exscientia’s goal to design all of its biologics de novo for specific target epitopes without the need for screening.
In order to design novel antibodies against specific protein epitopes, it is necessary to generate accurate models of their structure at speed and scale. Initial versions of the technology invented by Professor Charlotte Deane, Exscientia’s Chief Scientist of Biologics AI, produced accurate protein modelling up to 35,000 times faster than Alphafold21.
Key to Exscientia’s AI approach is using the knowledge of the observed human antibody space to optimise biologics for clinical development. The binding site of the antibody consists of two chains (heavy and light). Typically, sequencing of antibodies has been limited to single chains, losing the true biology of antibodies. The company is building a proprietary database of paired chain sequences to better understand the complex biology of antibodies in a more natural environment. Exscientia is using this data for machine learning in order to describe and model human antibody space more accurately.
Exscientia’s new laboratory facilities will automate the production of proprietary data for each antibody, measuring essential qualities including affinity, immunogenicity, aggregation and stability. Exscientia’s engineers are building proprietary automation hardware to enable high-throughput antibody profiling to support its predictive model building for multi-parameter optimisation.
Professor Andrew Hopkins, Exscientia’s founder and Chief Executive Officer, said: “Expanding into biologics enables us to nearly double the addressable target universe of our platform. Over the past two years, antibodies represented eighteen of the top fifty drugs measured by revenue, and by adding the capability to design new human antibodies with AI and automation, we believe we will be able to develop the most effective drug for patients, regardless of modality.”
“Our strategy is to replace current experimental discovery techniques with precision engineered de novo design of optimised, fully-human biologics,” said Professor Charlotte Deane MBE. “Current methods limit the discovery of new binding sites to what can be explored via animal immunisation or laboratory-based libraries. By virtually designing all aspects of a biologic in silico, we can explore a much broader target universe and create more precisely targeted therapeutics: Biologics by design not discovery, driven by AI and automated experiment, is our approach.”
- Abanades et al. Bioinformatics 2021