This presentation explores the integration of dynamics into the intrinsic physicochemical properties of antibody-based therapeutics with the aim to better understand and predict protein behavior in different environments and formulations.

The LNP-mRNA platform is generating considerable interest in the field of immunotherapy. To date, there are no specific regulatory guidelines for the identification and mitigation of unwanted immunogenicity risk factors for LNP-mRNA products. Here, we present a strategy utilizing a suite of in vitro immunogenicity/reactogenicity assays that could be applied early in drug discovery to guide the design and optimization of LNP-mRNA therapeutics to reduce immunogenicity risk factors.

Antigen presentation of MHC Class II plays an essential role in mediating the anti-drug response to large-molecule drugs. Such a response reduces drug efficacy and potentially causes safety concerns. Here, we develop graph-pMHC, a graph neural network approach to predict pMHCII presentation. We demonstrate that graph-pMHC dramatically outperforms other methods, such as NetMHCIIpan-4.0 (+22.84% average precision). We further create an antibody drug immunogenicity dataset from clinical trial data, and develop a method for measuring anti-antibody immunogenicity risk using pMHCII presentation models. Our approach outperforms Biophi’s Sapeins score by 7.14% ROCAUC for predicting antibody drug immunogenicity.

We report a high-throughput protein engineering method for rapidly identifying antibody candidates with both low self-association and high affinity. We conjugate IgGs that strongly self-associate to quantum dots and use these conjugates to enrich yeast-displayed antibody libraries for variants with low levels of immunoconjugate binding. Deep sequencing and machine learning analysis enables identification of extremely rare variants with co-optimized levels of low self-association and high affinity.

The presentation will focus on a new end-to-end high-throughput biologics engineering platform. It describes the generation and multiparameter characterization of large panels of biological molecules enabling short design and learning cycles. Here, we report on how we apply this new high-throughput engineering platform for parallel multiparametric optimization of protein therapeutics and how these high-quality datasets can be applied for machine learning applications.

Sanofi's automated high-throughput engineering platform enables the rapid generation of large panels of multi-specific antibody variants, resulting in the accumulation of big data sets. By mining these data sets, we were able to extract engineering patterns and develop AI-based virtual screening workflows to guide the exploration of vast design spaces in biologics drug discovery.

Developability assessment of drugs during the discovery phase is critical to ensure the manufacturability, safety, and efficacy of selected candidates and thus improve the likelihood of clinical success. In this presentation, I will describe the developability framework at Large Molecule Research, pRED, Roche, with a special focus on assessment of molecule suitability for high concentration formulations and automation approaches for high-throughput developability analysis.

For an antibody to be a successful therapeutic candidate many competing factors must be optimised simultaneously including desired binding affinities, good biophysical characteristics, and low immunogenicity. Here we will discuss the development of interpretable, biophysically-meaningful, deep-learning predictive models to optimised viscosity and other developability properties to accelerate the discovery and development process. These methods, along with high-throughput screening allow for rapid identification of lead molecules with good biophysical characteristics.