How artificial intelligence is reengineering protein engineering.

Over the past decades, protein engineering has matured into a field of its own, driven by computational modeling and high-throughput wet lab experiments, with broad application in therapeutics, diagnostics, agriculture, and manufacturing. In recent years, artificial intelligence (AI) has further propelled protein engineering by enabling more efficient search through high-dimensional sequence space for proteins with desired properties. Notable AI-based advances encompass generative modeling of sequences, backbone structure, and atoms; tailoring general versions of such models to design proteins with specific properties; modeling for extraction of protein representations and scoring candidate protein sequences; and developing techniques for library design, including synthesis-aware approaches. Herein we discuss these advances, emphasizing a unifying view through a statistical interpretation of modern AI approaches. Editor's summary: Proteins, with their varied structure and chemistry, are the prime actors of biology and have long been targets for in vitro and in silico engineering. Generative protein models and other artificial intelligence (AI) tools are now being integrated into experimental workflows. Listgarten and Jiang reviewed advances in AI methods and discuss how statistical principles are being used to transform protein engineering through conditional generative modeling. Beyond the sizable advances so far, current challenges include designing functional enzymes, disordered proteins, and binders of all kinds, problems for which we currently lack sufficient training data. —Michael A. Funk [ABSTRACT FROM AUTHOR]