I present deep screening, an ultra-high-throughput approach leveraging the Illumina HiSeq platform for massively parallel sequencing, display, and rapid affinity screening at the level of >10e8 individual antibody-antigen interactions. Deep screening enabled the discovery of mid- to high-picomolar single-chain Fv (scFv) antibody leads directly from unselected, synthetic scFv repertoires in a three-day experiment, and provides large sequence/function correlation datasets suitable for machine learning. Used as an input for a large language model, deep screening data allowed the de novo generation of scFv sequences not present in and with higher target antigen affinity than the original library.

Antibody somatic variants can be mined from immune repertoires to rapidly optimize the affinity of antibodies derived from immunization. Antibody genetics interpreted in the context of functional data can also be used to more broadly mine repertoires for unrelated clones that share epitope specificity for applications beyond affinity optimization.

Despite their importance in research, monoclonal antibodies are not systematically sequenced. We developed Nanopore Antibody Sequencing (NAb-seq), a three-day, species-independent, and cost-effective workflow to characterize paired, full-length light- and heavy-chain genes from hybridomas. NAb-seq is accurate, scalable, and can identify multiple heavy- and light-chains within a single cell line. We further show that NAb-seq is applicable to single cells, allowing antibody discovery in rare populations such as memory B cells.

Recent advancements in deep sequencing and microfluidics now enables the high-throughput recovery of paired heavy- and light-chain sequences at single-cell resolution. I will discuss how single-cell immune repertoire sequencing can be used to discover antibodies against model antigens and therapeutic targets, and how computational features such as clonal expansion, transcriptional phenotypes, and antibody evolution relate to biophysical properties such as specificity, affinity, and epitope.

In the discovery of antibodies as therapeutics, traditional methods for screening and optimizing antibodies are limited in the sequence space complexity they can cover and are generally resource intensive. We discuss approaches that combine advances in in vitro display and computational deep learning to enable more efficient searching of antibody sequence space. We also discuss how integrating sequence and structural information using deep learning enables improved prediction of antibody properties

Accurate prediction of biophysical properties is an important but challenging problem in antibody therapeutic development. To address the limitations of small training datasets, we constructed a novel hybrid machine learning and protein modeling framework. This framework utilizes structure-derived molecular descriptors, sequence embeddings from protein language models, or 3D geometric representations of structures. Our findings indicate that molecular descriptor-based models trained within an active-learning framework provide superior and more interpretable predictions.

• Identifying the highest priority opportunities for AI in biologics discovery, alongside the primary blockers and risks that currently stifle progress
• Strategies for enhancing data diversity, access, preparation and management to increase the accuracy of AI models
• Speculating on the next significant breakthroughs that could revolutionize biologics discovery
• Assessing the impact of partnerships and collaborative projects on the advancement of AI in biologics discovery, including the hurdles these collaborations face​

Agonistic antibodies directed to immunostimulatory receptors are an untapped source for immunotherapy. Here we discuss the properties required to optimally agonize these receptors and describe potential strategies for leveraging them for immune activation and anti-tumor efficacy. Using TNFR superfamily receptors as a paradigm and IgSF members for comparison, we evaluate the key characteristics of the Fc, hinge, and Fab in delivering powerful receptor agonism. Specifically, we explore the impact of FcgR engagement, epitope, hinge flexibility, and F(ab) affinity in modulating agonistic activity for multiple mAb and receptor types.

Fibroblast growth factor receptor 1 (FGFR1) is a promising yet challenging therapeutic target that may require an agonist or an antagonist depending on the indication. We investigated the mechanism of action of reported agonistic and antagonistic FGFR1 antibodies and described a novel, functionally-distinct FGFR1-active conformation that impacts in vivo activity. We further engineered these antibodies and demonstrated that modulating the geometry of FGFR1 can effectively change the signaling outputs.

Targeting the brain with biologics has proven challenging due to the restrictions of the blood-brain barrier (BBB). Here, we introduce a high-throughput in vivo screening based on Manifold's mCode technology to evaluate BBB uptake across over 1,000 antibodies targeting 8 different receptors. Our approach reveals antibodies with diverse phenotypic characteristics and pharmacokinetic profiles with increased brain biodistribution. We leverage this large dataset using machine learning to further engineer these antibodies and enhance their brain uptake and residence time. Validated in murine and primate models, our findings enhance BBB transcytosis understanding and mark a shift in early drug discovery methodologies.

In this talk we’ll cover a novel generative AI and experimental screening technology for designing selective biologics for multipass membrane protein targets. We’ll describe the development of AI technologies to generate fully genetically encodable, stable, and soluble proxies of an ion channel and GPCR that structurally mimic the native multipass receptor. These proxies can be used to screen and identify antibodies and other biologic binders from AI or randomly generated libraries that bind the native membrane-bound receptor. This represents a new approach to make some of the most challenging and high-impact drug targets more tractable for drug discovery.

Complex (challenging) targets are now becoming the norm for therapeutic antibody discovery. The availability of new immunogens (genetic, protein, and cellular) combined with novel approaches for immunogen delivery, and novel use of adjuvants can make challenging targets less challenging. Here we show our advances in improving the immune response to challenging targets that increase the probability of isolating therapeutic antibodies in our antibody discovery process.