From chemical space to synthetic reality: a framework for makeable-by-design generative AI in drug discovery

Abstract

Generative artificial intelligence (AI) is reshaping molecular design, but many outputs are still judged by proxy metrics that do not establish whether proposed compounds can be synthesized, tested and advanced. This review reframes the field around ‘synthetic reality’: the extent to which AI-generated molecules survive route planning, precursor availability, experimental execution and medicinal chemistry decision-making. We examine the limits of heuristic synthesizability metrics, assess route-aware and execution-facing systems and propose maturity levels, reporting standards, actionability metrics and decision gates for auditable real-world evaluation. Future progress will depend on shifting from molecule generation alone to experimentally actionable design.