Artificial intelligence (AI) has reshaped protein design by enabling models trained on large-scale sequence and structure data to generate proteins with specified functions. These models are best understood in the context of an end-to-end pipeline that includes data curation, model development, candidate generation and filtering, and experimental validation. Here, we review AI-driven protein design methods that span this full pipeline. We begin with a primer on AI-driven protein design and then outline the key components of the pipeline and assess performance across three major application areas: binders, antibodies, and enzymes. By consolidating experimental outcomes across diverse approaches, we provide a practical reference for methods that currently succeed in the lab and highlight the ongoing importance of experimental feedback in advancing AI-driven protein design.
Kosonocky et al. (Sun,) studied this question.