Targets for biologic drugs, and drug modalities themselves, are becoming more and more complex. In addition, increasing focus is being placed on drug mechanism-of-action. Due to this advancement in biologic drug technology, it has become critical to include assays measuring complex functional activity higher in the screening cascade. I will describe how we have been developing and implementing these high-throughput functional assays to screen for complex biology.

We evaluated agonist antibody discovery rates in binding-biased and unbiased libraries using Triplebar's Hyper-Throughput Screening system (HyTS). Employing a microfluidics-based paracrine discovery platform, we sorted antibody-secreting cells based on immune cell responses. Our investigation focused on identifying functional Abs and exploring the benefits of unbiased searches for novel agonists.

BigHat Biosciences has developed novel machine learning (ML) approaches that leverage our high-speed, automated wet lab in order to rapidly and iteratively design hundreds of next-generation therapeutic antibodies each week. BigHat’s algorithmic approach pairs with our unique wet lab to guide the search for better molecules by learning from each cycle of characterization across multi-objective affinity, function, and developability measures of each antibody. We’ll discuss several methodological developments in multi-parameter optimization, active learning (Bayesian Optimization), and generative humanization, and highlight the functional and in vivo validation of our designs.

IdeS from Streptococcus pyogenes cleaves human IgG subclasses, severing the antigen binding domains from the Fc that mediates immune effector functions. IdeS is used clinically for desensitization of kidney transplant recipients and may have applications in autoimmunity and gene therapy, but immunogenicity prevents repeat dosing. Using computational algorithms, antigenic epitopes for CD4+ T cells and B cells were removed, while achieving high IgG proteolytic activity and specificity with improved pharmacokinetics.

Antibody discovery has made rapid progress against simple targets like soluble ectodomains, but discovery remains difficult against challenging targets like expanded viral families and membrane proteins. Here, we will share recent case studies and unpublished data for miniaturized antibody high-throughput screening against difficult targets, including to discover functional antibodies against infectious disease antigens, and against membrane proteins.

Historically, biologics discovery has primarily been an experimentally-driven enterprise. Yet, the past decade has seen remarkable progress in computational protein design, structure prediction, and machine learning methods. We leverage these technologies to accelerate the biologics discovery process, as well as to design the next-generation multispecific antibodies to enable new biologies.

T-cell epitope sequencing (or Tope-seq) is a high-throughput screening platform enabling rapid, in vitro function-based assessment of T cell receptors (TCRs) against up to a million DNA-coded peptide sequences simultaneously. Using the Tope-seq pipeline, along with our proprietary human whole proteome-coding minigene library and engineered effector/target chassis systems, we are currently applying our platform of technologies towards interrogating potential autoimmune cross-reactivities in candidate TCR therapeutics, de-risking their future clinical development.

Peptide-HLA (pHLA)-targeting therapeutics can drive T cell killing of target cells based on altered intracellular protein expression. Among pHLA-targeting modalities, soluble T cell receptors (TCRs) have evolutionarily engrained advantages in peptide specificity yet require affinity maturation for therapeutic efficacy. To overcome this barrier, we developed a novel yeast-based platform that rapidly generates high-affinity TCR variants that elicit potent T cell activity in vitro, accelerating the development of soluble TCR-based therapeutics.

The blood-brain barrier (BBB) restricts the transport of large molecules between the blood and brain tissue, posing a challenge for the delivery of therapeutics to the brain. Fc-based transport vehicles (TVs) are a novel approach to brain delivery that exploit receptor-mediated transcytosis to transport biotherapeutics across the BBB. In this presentation, we will discuss the development of TVs and their potential for brain delivery of therapeutic proteins.

Prescient Design, a Genentech accelerator, is developing novel computational tools for optimizing antibody affinity and multiple developability parameters by combining ideas from machine learning and structural biology. In this talk, I will give an overview of our lab-in-the-loop framework that consists of our novel generative modeling approaches, combined with multi-objective optimization, and active learning framework.

The emergence of ML-enabled technology platforms that aim to enhance molecule performance have the potential to revolutionize the way we approach drug discovery. However, without a purpose-built tech stack that puts data quality at the heart, many are destined to fail. This talk will focus on the deep integration of predictive assays, data generation, data capturing, and data pre-processing needed to enable iterative active learning cycles for lead optimization.

Antibodies are an important class of medicines whose efficacy is driven by specific target binding. Given the therapeutic relevance, there have been multiple attempts to computationally predict how mutations affect binding affinity. Using experimental and synthetic data, we demonstrate that there is currently not enough experimental data available—by orders of magnitude—for accurate, generalizable prediction. We also investigate the role of diversity and suggest guidelines for robust machine learning model development.

We present an innovative system integrating microfluidics and machine learning for high-throughput immunotherapeutic drug discovery. Our approach aims to decipher molecular design principles by effectively screening and analyzing large libraries of immunotherapeutic candidates directly for their end function (e.g., T cell activation). The system combines a cutting-edge microfluidic platform for rapid and precise experimentation with machine learning algorithms to predict and optimize candidate performance. This integrated approach holds great promise for accelerating the development of effective immunotherapeutic drugs.

The AWSEM (Associative memory, Water-mediated, Structure, and Energy Model) coarse-grained molecular dynamics model is an advanced computational tool that has found utility in several use cases characterizing the molecular dynamics of proteins, including the prediction of  binding interfaces of select homodimers and heterodimers (Zheng et al. 2012), intrinsically disordered proteins (Wu et al. 2018), and high-throughput TCR-pMHC repertoire characterization (Lin et al. 2021). We are energized at the prospect of modernizing and building on this rich body of technology innovation. Here we describe our work at Deep Origin in productionizing this powerful model towards the application in dimeric protein engineering use cases including antibody development, including its internal deployment in our development organization.