In the case study presented here, a therapeutic grade highly potent anti-IL-5 antibody was significantly improved through combining a yeast-based antibody selection platform with a highly efficient affinity maturation method, using structure based as well as unbiased CDR targeted diversification. CDR-targeted mutagenesis libraries were constructed using the Adimab Yeast Platform and antibodies with improved affinity for IL-5 were selected by fluorescence activated cell sorting. This combined approach enabled the identification of sub-pM affinity range antibodies using a single rapid maturation cycle.

• NEW DATA - This Presentation Contains New Data
• NEW TALK - This Presentation Will be Given for the First Time

First PK profiling of multispecific antibodies with various backbones and Fc mutations in huFcRn transgenic mice compared to Non-Human Primates. Tg32 mouse model can be used to distinguish PK differences between antibodies containing Fc-backbone mutations in the same range of order as observed in NHP. This model may serve as a convenient, cost effective surrogate for in vivo prediction of antibodies t_1/2z, Vss and Clearance facilitating the screening strategy to select antibodies with the most favorable PK. By extension, a first estimation of PK parameters in human can be directly deduced from this mouse model.

Typically, immunoglobulin G antibodies depend on homodimerization of the fragment crystallizable regions of two identical heavy chains. By modifying the CH3-CH3 interface, with different mutations on each domain, the engineered Fc fragments form rather heterodimers than homodimers. Here, we use classical molecular dynamics simulations to identify key interactions in the CH3-CH3 and CH1-CL interfaces, that contribute to their stability and tendency to heterodimerize. Additionally, we compare CH3-CH3 domains, to the structurally similar CH1-CL interfaces in antigen-binding fragments (Fabs). The CH1-CL domain shares a very similar fold and interdomain orientation with the CH3-CH3 dimer. Thus, numerous well established optimisation efforts for CH3-CH3 interfaces, have also been applied to CH1-CL dimers to reduce the number of mispairings in the Fabs. By comparing all CH3-CH3 and CH1-CL dimers with each other we find similar interaction patterns. However, we also observe pairing specific interface interactions and show that also different combinations of heavy chains with k and l-light chains result in characteristic interface contacts. Additionally, we present exemplary dissociation mechanisms for both CH3-CH3 and CH1-CL dimers and identify key interactions that contribute to stability and their tendency to heterodimerize. Thus, this study has broad implications for improving the yield of bispecifics as it provides a structural and mechanistical understanding of CH3-CH3 and CH1-CL interfaces and thereby presents a crucial aspect for the development and optimisation of bispecific antibodies.

Computational methods can accelerate the selection of therapeutic protein candidates with suitable biophysical properties and guide formulation development. Here, we will present approaches for the selection of aggregation-resistant proteins and the identification of new stabilizing excipients in silico. Furthermore, we will show how molecular dynamics simulations can provide mechanistic explanations of buffer effects and stabilizing protein-excipient interactions. We will conclude by discussing the opportunities and limitations of the presented approaches.

• NEW TALK - This Presentation Will be Given for the First Time

We will discuss the challenges in developing a neutralizing anti-IL-21R antibody, that can effectively compete with IL-21 for its highly negatively-charged paratope, while maintaining favorable biophysical properties. To overcome the limitations in charge-based in vitro deselection methods, we utilized a combination of structure-guided rational library design, next-generation sequencing of library outputs and machine learning methods to identify a high-affinity antibody with desirable stability and biophysical properties. We have expanded upon these methods to develop additional predictive tools to improve the biophysical properties of antibodies.

The Specifica Generation 3 Library Platform is based on highly developable clinical scaffolds, into which natural CDRs purged of sequence liabilities have been embedded. The platform uses phage+yeast display to directly yield highly diverse (100-1000 clusters differing by Levenshtein distance 30-40), high affinity (20% subnanomolar), extremely developable (>80% lack biophysical liabilities), drug-like antibodies, which in a recent Covid campaign were as potent as antibodies from immune sources.

An efficient solution for the correct assembly of IgG-like bispecific antibodies is a sought-after component for the manufacturing of a growing number of therapeutic antibodies entering the clinic.

Using a panel of bispecific antibodies we will present a  comprehensive evaluation of the “Lonza bispecific platform”, showing that this technology provides an excellent, easy-to implement solution for the manufacturing of bispecific IgG-like antibodies composed of four different chains.

There is an increasing interest in developing computational methods to predict the developability of antibodies in order to reduce the cost of antibody discovery programmes as well as late stage failures. Among the major causes of attrition in these programmes, poor specificity stands out, in part because it is challenging to assess it experimentally in a high-throughput manner. The availability of large databases of antibody sequences and data about antibody specificity is making it possible to develop computational methods to predict the specificity of antibodies from their amino acid sequences. I will present a novel method to achieve this goal.

The current race to develop better drugs faster has led biopharmaceutical companies into optimizing all drug discovery and development processes. As part of this effort, machine learning algorithms are being developed to identify correlations between amino acid sequences and physicochemical properties observed during drug development. Here, we describe our ongoing efforts aiming at benchmarking such machine learning algorithms to facilitate our drug discovery process. The audience will learn about Merck MSD current strategy in order to enable in silico approaches for developability assessments. They will learn about useful experimental assays endpoints for in silico approaches, data collection, curation and usage as well as the new platform Merck MSD is currently developing to serve as the interface bewteen the Scientists and machine learning algorythms for in silico developability assessments.

• NEW DATA - This Presentation Contains New Data
• NEW TALK - This Presentation Will be Given for the First Time

GenScript precision mutant libraries by electrochemical semiconductor DNA synthesis technology allow user-defined codon with precise ratios, saving valuable time and effort during the screening and characterization process, speeding up the engineering workflow, and reducing the overall cost of downstream expenses. We showcase a case study of our precision mutant libraries in affinity maturation of a monoclonal antibody with a  ~10,000 fold improved affinity via site saturation and combinatorial mutant library.

Advances in artificial intelligence have demonstrated its ability to drive and accelerate new scientific discoveries. We show that the complex sequence structure relation can be learnt efficiently using deep neural networks trained on crystallized protein structures. We attempt addressing two protein modelling problems here: protein side-chain geometry prediction and predicting stabilizing protein sequences. These models achieve improved efficiency when compared against the state-of-the-art protein packing and protein design tools respectively.

• ​NEW TALK - This Presentation will be given for the First Time

We recently collected amino acid sequences of currently marketed antibody-based biologic medicines and used these to derive their homology-based structural models. Availability of both sequence and structural models of these biotherapeutics afforded us an opportunity to analyze their physicochemical attributes from a perspective of developability. These analyses have resulted in a profile that can be used to estimate ‘medicine-likeness’ of the biologic drug candidates currently in discovery and development.

Since the Leap-In Transposase platform was launched for biotherapeutic cell line development three years ago, the market adoption has been robust, and it has rapidly become well accepted in the industry.  This talk will highlight key milestones from these past few years and will focus on case studies around our COVID response, chain ratio balancing for complex molecules and how the technology can be used to reduce target gene expression.