Staphylococcus (S.) aureus infection is a frequent cause of sepsis in humans, a disease associated with high mortality and exacerbated by antibiotic resistant strains. The development of immune therapeutics against S. aureus has been thwarted by the formidable array of virulence mechanisms of this pathogen that together inhibit antibody's functions and subvert host immune defenses. Here, we describe engineered antibodies that escape bacterial-mediated inhibition, display extended half-life and improve bacterial killing.

The inability of diverse biomolecules to readily penetrate the blood-brain barrier is a key limitation to their use in research, diagnostic, and therapeutic applications. We are developing bispecific antibodies that engage either CD98hc or transferrin receptors and efficiently transport biomolecules into the CNS. We will discuss the unique advantages of each shuttling pathway, our progress in developing second-generation shuttles, and their drug delivery applications.

Glucose-dependent insulinotropic polypeptide (GIP) plays a major physiological role in nutrient deposition and storage. GIP is overexpressed in obese individuals, and enhanced postprandial secretion initiates a vicious cycle characterized by increased nutrient uptake and storage in adipocytes, leading to insulin resistance and hyperinsulinemia, further increasing adipocyte nutrient uptake and storage. A reduction in GIP signaling using a specific monoclonal antibody to GIP decreases weight in obese mice and prevents weight gain in normal mice and in leptin-deficient ob/ob mice without suppressing food intake, suggesting that this antibody might provide a novel, useful method for the treatment and prevention of obesity.

De novo designed miniproteins represent a groundbreaking new modality, offering unprecedented flexibility in engineering biologics with an array of desirable features tailored for specific applications. We developed a novel GLP1R agonist miniprotein designed  de novo to exhibit improved biased signaling with a strong emphasis on G protein-coupled signaling, while significantly reducing ß-arrestin signaling. Our well-folded and stable agonist shows blood glucose lowering in diabetic db/db mice.

Advances in AI/ML are transforming biologics drug design by optimizing antibody engineering, enhancing prediction of protein structures, and accelerating discovery timelines. This presentation will explore recent breakthroughs and real-world applications of machine learning in designing innovative biologics, with a focus on antibodies. The discussion will also highlight emerging trends, future possibilities, and the potential of AI-driven approaches to overcome complex challenges in engineering antibodies for diverse therapeutic applications.

We discuss herein the utility of deep learning in capturing the latent embedding space to describe long-range interaction, biophysical properties, and downstream prediction task. The value of combining deep and shallow learning to operationalize AI for antibody design. And, we exemplify the application AI for antibody design by combining LLM embedding and shallow machine learning to predict the expression level of mAb and VHH in mammalian cells.

Antibodies are versatile therapeutic molecules that utilize immense combinatorial sequence diversity to cover a vast fitness landscape. However, designing optimal antibody sequences remains a major challenge. We introduce AlphaBind, a transformer-based model pre-trained on millions of antibody-antigen affinity measurements. In four parental antibody systems, we show that AlphaBind accurately predicts affinity using just one round of unguided data generation, allowing effective sequence optimization while preserving diversity and key biophysical properties.

High specificity is a key attribute of biologic drugs. Screening for high levels of non-specific binding (NSB) is an important part of assessing the overall developability profile of biologics candidates. I will present a machine learning workflow for predicting BVP—one of the most used NSB assays. Using descriptors based on computed protein structures, the workflow identifies several risk factors which tend to correlate with increase in NSB. The workflow is purely in silico and can dramatically reduce the cost and increase the throughput of early NSB screening.

We have advanced multivalent miniprotein engineering platforms to achieve selective control of target engagement in a size-efficient manner. We will present the effects of different molecular formats for multivalent binding, including the impact of paratope linkages and monovalent affinity on resultant selectivity and utility across multiple applications.

Infection by human cytomegalovirus remains a concern for newborns and immunocompromised individuals. No vaccines or antibodies are available, but adoptive T cell therapy has shown promise. Accordingly, we engineered a human T cell receptor to have high affinity and peptide specificity for the immunodominant pp65/A2 complex on infected cells. After reformatting as a bispecific T cell engager, this protein potently activated T cells against peptide-pulsed and infected fibroblasts. 

We are developing a strategy for antibody design that addresses antibody affinity and developability by combining atomistic design and machine-learning calculations. Our work demonstrates that affinity can be enhanced by designing mutations that stabilize light-heavy chain interfaces or the CDRs, and that these enhancements often correlate with improvements in stability and antibody expressibility. I will also discuss new work to design universal antibody repertoires for accelerated discovery of developable antibodies.