Abstract
Generative artificial intelligence is reshaping how researchers discover protein-binding compounds and develop them into drug candidates. Unlike traditional methods that screen existing molecules, structure-based generative AI designs novel compounds tailored to a protein's three-dimensional binding pocket. This review outlines how these approaches are applied in early drug discovery, focusing on general principles. We categorize methods according to their generative modeling paradigms and their strategies for using structural data to guide molecular design, distinguishing de novo incremental builders from models that generate full structures. We also survey lead-optimization techniques, highlighting a recent shift toward generation-driven medicinal chemistry.