Multi-objective antibody design and optimization using machine learning

Abstract

Antibodies, ranging from conventional immunoglobulins to single-domain and bispecific antibodies, are developed for a wide spectrum of therapeutic and diagnostic applications. However, lead antibodies are commonly engineered to optimize one or more key properties, including affinity, specificity, polyreactivity, stability, aggregation and immunogenicity, and trade-offs among these attributes frequently hinder the development of variants suitable for clinical use. Furthermore, identifying lead candidates with desirable properties typically requires multiple rounds of experimental screening and evaluation, particularly when targeting pre-selected epitopes or challenging, high-value targets such as G protein-coupled receptors (GPCRs), ion channels and multipass membrane proteins. Artificial intelligence (AI) and machine learning (ML) methods can streamline experimental screening and antibody generation methods. In this Review, we discuss both computational multi-objective optimization of existing antibodies and de novo antibody design. The application of AI and ML methods has the potential to overcome current bottlenecks in antibody development and accelerate the generation of improved therapeutics for human health. Antibodies are among the best-selling drugs and are crucial for numerous therapeutic and diagnostic purposes. However, optimizing antibodies for desirable drug-like properties has long been a hurdle, delaying or preventing promising drugs from reaching patients. Advances in computational methods, especially those using artificial intelligence and machine learning, can streamline and enhance conventional methods, enabling more efficient discovery and optimization of therapeutic antibodies.