Although antibody drugs have great therapeutic benefits, most function primarily as inert antagonists. They don't fulfill the potential of proteins, which are dynamic machines that can sense changes in the environment and respond to them. Biolojic Design's clinically validated AI-driven platform designs dynamic antibodies that are programmed to respond to changes in the environment The first such antibody is in phase 2. I will describe Biolojic’s platform, which experimentally generates data for AI models. I will focus on the computational aspects, which aim to optimize, simultaneously, multi-antibody properties, which in turn allow the design of smart and dynamic antibodies.

AI/ML applications to the field of protein engineering have recently generated immense enthusiasm. In addition to protein folding and de novo design, statistical and deep learning technologies are revolutionizing many aspects of the biologics therapeutic discovery. We cover some of the real-world use cases for AI/ML applications—from hit identification to lead optimization stages of discovery biologics—that are facilitating the computational identification of high-quality preclinical candidates and improving efficiency.

Exploring the bilateral importance of leveraging strategic data collection and intelligent experimental design to enhance AI capabilities in antibody development. Discussing the utilization of evolutionary intelligence to refine and optimize affinity maturation strategies, leading to the discovery of antibodies with superior binding properties and therapeutic potential.

The path from discovered binder to developable modern antibody therapeutics represents a complex protein engineering challenge. At BigHat we frame this as a multi-objective Bayesian optimization problem, with machine learning models in-the-loop with a high-throughput wet lab on a weekly build-test-train cycle. We present several case studies of novel predictive and generative ML methods deployed on this platform to solve real-world antibody therapeutic engineering challenges.

Structure-based vaccine design campaigns that aim to stabilize full-length proteins will not succeed when the virus is evading immune response by sequence diversity. In this work, we capitalized on the recent major advanced that have been achieved in protein design using generative AI tools. We in silico designed de novo proteins that scaffold sequence-conserved epitope regions of the antigens of interest. Next, we incorporated the in silico designed mini proteins onto self-assembling nanoparticles and performed pre-clinical animal immunization studies eliciting immune responses.

The design and discovery of early-stage antibody therapeutics is time- and cost-intensive. I will present an end-to-end machine learning-driven single-chain variable fragments (scFv) design framework that uniquely combines large language models, Bayesian optimization, and high-throughput experimentation. The method enables rapid and cost-effective design of thousands of scFvs across all complementary determining regions. The designed antibodies exhibit strong binding affinities, at high levels of diversity, to a given antigen.

This talk will cover how large language models of protein sequences and structures can learn evolutionary rules that help guide the artificial evolution of human antibodies. We will cover how algorithms known as protein language models can guide the affinity maturation of antibodies against diverse antigens using sequence information alone. Next, we will cover how multimodal language models can further improve the ability to guide antibody evolution completely unsupervised by incorporating information about the protein's structure, which we use to make clinical antibodies more potent and resilient against viral escape variants.

• What is the therapeutic potential of conditional activity and how do we best balance this against increased complexity and risk?
• What challenges are unique to ML-driven design of conditional molecules?
• How does the optimal ML toolkit vary between conditional and unconditional design?
• How best can we overcome challenges in data acquisition and availability?
• What are the currently tractable forms of conditional activity and what can we envision for the future?​

DeepAb, a deep learning model for predicting antibody Fv structure directly from sequence, was used in conjunction with experimental deep mutational scanning (DMS) enrichment data to design 200 potentially optimized variants of an anti-hen egg lysozyme (HEL) antibody. We discuss the improvement rate of the designed clones for affinity, thermostability, and developability—and what the results would have been without using experimental DMS data to guide the design process.

Optimal clone selection strikes a balance between selecting desirable antibody properties and diversity, aiming to maximize the range of targeted epitopes while minimizing redundancy. At Sanofi, we are exploring in silico screening, leveraging AI/ML methods to enhance success rates and accelerate the transition from discovery to lead. In parallel, generative AI can be used to augment natural and experimental repertoires, eliminating undesirable traits while preserving desirable characteristics.