Effective drug development and discovery relies on the availability and understanding of target proteins. The Structural Genomics Consortium (SGC) is at the forefront of advancing protein-based therapeutics by optimizing protein expression and characterization. We achieve 10 to 100 mg yields for high-throughput screening using refined constructs and various expression systems, including E. coli, insect, and mammalian cells. Rigorous testing and characterization enhance our understanding of biochemical and molecular mechanism of these proteins. This comprehensive approach supports the accelerated development of effective therapeutics, advancing the field of protein-based drug discovery and contributing to more rapid responses to emerging health challenges.

Antibody development involves complex characterization steps, which are time-consuming and limited by low-throughput assays. This presentation introduces microplate-based high-throughput systems for expressing and analyzing recombinant antibodies. Utilizing nanopore sequencing, surface plasmon resonance (SPR) for interaction analysis, and differential scanning fluorimetry (DSF) for thermal stability analysis, these systems enable efficient evaluation of antibody affinity, specificity, and stability, accelerating data-driven antibody design. Examples of antibody design using this system will be presented.

Chimeric antigen receptor (CAR) T cells have transformed cancer treatment in the clinic, making a significant impact in the treatment of B cell malignancies. As we search for the next therapy, the need to screen large, architecturally diverse CAR libraries requires a robust high-throughput manufacturing process. Here we describe a novel high-throughput CAR T cell manufacturing and screening process for the identification of tumor-specific antigen CAR T cells.

Bispecific discovery can be accelerated through implementation of an efficient bispecific production platform. Through generating fit-for-purpose bispecifics ready for high-throughput functional assessments, timelines from target identification through lead optimization can be accelerated. Case studies from preclinical programs demonstrate the ability to rapidly identify preferred therapeutic formats and optimize bispecifics as well as efficiently interrogate large bispecific sequence spaces though leveraging sampling methods.

Designing proteins with desired functionality is a fundamental challenge of protein engineering. I'll be sharing our team's progress towards combining high-throughput experimental assays and deep probabilistic modeling to engineer proteins with user-defined properties.

As biologic therapeutics continue to increase in complexity, innovative approaches to candidate screening, production, characterization, and development are more important than ever. Our advanced protein production workflows incorporate novel processes, intelligent high-throughput automation, and high-quality informatics to enable robust molecule screening, selection, and scale-up. These enhancements enable advances in the speed, quality, and productivity of our biologics development pipeline.

The complexity of mammalian cell culture and the heterogeneity of large-molecule products have historically limited the application of robotic automation platforms in production and characterization to mainly either early stages for small-quantity, stage-gate quality material, or later stages for industrializing specific task accomplishments. Here, we reveal our next-generation automation strategy to bridge the gap and prepare large quantities of high-quality material, solving an essential need for more complex biologics modalities.

• How do we track data captured at every stage of the protein production lifecycle?
• Is higher throughput ALWAYS better or does it just push the bottleneck downstream?
• How do we connect, analyze, and correlate data from different sources (instruments, databases, e-notebooks) in order to draw conclusions from our data?
• How do we effectively QC HT data?
• How can we use our data to reduce the number of samples in future design-test-make cycles?

• HT production approaches
• Incorporating AI/ML predictions into workflows
• Challenges with emerging modalities
• Functional assessment of molecules

Recombinant protein expression and production is a critical step in drug discovery-and secreted proteins have been a rich source of therapeutics and drug targets. To support growing demand for these proteins, we developed methods to triage protein expression constructs effectively by automating processes, increasing capacity, and reducing costs. This presentation focuses on small-scale high-throughput secreted protein expression platform and the crucial collaboration between different sub-groups to provide efficient results.

The production of highly purified, well-characterized protein reagents for structural biology generates large and diverse datasets, spread throughout various instruments, notebooks, and databases. In our group, we have developed an optimized HT expression workflow that utilizes digitized, mineable data wherever possible. This, coupled with data centralization and visualization through various dashboards, enables us to more accurately predict protein behavior based on construct design.

Multispecific antibodies (msAbs) are increasingly becoming the preferred format due to their ability to modulate a wide range of biological targets. However, msAbs present unique protein production challenges due to product-related impurities that are difficult to remove without losing the protein of interest. Combined purification-enabling mutations (PEMs) and charge-pair mutations (CPMs) have been shown to enforce the correct chain pairing of msAbs and their productivity. This combination could accommodate a wide range of production scales including medium- to high-throughput purification workflows for msAb.

Accurately predicting interactions between small molecules and proteins is an unsolved problem that will require large datasets with excellent fidelity. At Leash, we're building infrastructure to design, execute, and analyze DNA Encoded Chemical Library data at-scale with a small team. During this talk we will present some of the problems and solutions we've developed over the past year that allow us to rapidly visualize billions of data points.

This presentation will focus on the convergence of a new high-throughput antibody discovery platform capable of screening 100s of millions of antibodies with machine learning to accelerate the full discovery process. This work is resulting in the identification of high affinity, developable modalities fit for therapeutic use in accelerated time frames while generating significant amounts of data-further refining our algorithms and models.

Protein language models (PLMs) generate high-dimensional data, creating challenges in storage and analysis. We develop abstractions to manage this complexity, focusing on key attention patterns to reduce storage and computational demands while retaining essential information. (1) Efficient data management uses abstractions and manages high-dimensional attention matrices, enabling scalable analysis of full proteomes. (2) We identify protein functions by finding key attention patterns associated with protein families, preserving relevant details within the compressed data. This approach improves precision and efficiency in protein function prediction, particularly for unannotated sequences, by optimizing computational resources and enhancing the scalability of PLM applications in protein science.