Artificial-Intelligence-Driven Design and Digital Execution of Chemical Reactions: A Review of Modern Software Platforms and Autonomous Workflows

The practice of chemical synthesis is being transformed by the integration of artificial intelligence (AI) with laboratory digitalisation. Where the design of a synthetic route once depended almost entirely on the tacit knowledge of an expert chemist, data-driven models can now propose retrosynthetic pathways, predict reaction outcomes and recommend optimal conditions, while robotic and continuous-flow platforms execute the resulting protocols with minimal human intervention. This article reviews the state of the art across the full pipeline — from AI-based reaction design to digital execution — and analyses the modern software ecosystem that makes it possible, including RDKit, IBM RXN for Chemistry, ASKCOS, AiZynthFinder, Synthia (Chematica), the Open Reaction Database and the Chemputer/XDL chemical-programming framework. Drawing on a structured review of 198 studies, we show that reaction-outcome prediction accuracy has risen from 57% for rule-based expert systems to over 90% for recent language-model architectures, that AI-guided Bayesian optimisation reaches near-optimal yields in roughly an order of magnitude fewer experiments than grid search, and that autonomous "self-driving" laboratories increase experimental throughput by 15–20 fold while cutting reagent consumption by more than half. We discuss application perspectives across pharmaceutical, materials and polymer chemistry, and outline the data, reproducibility and governance challenges that must be resolved for these methods to be adopted broadly and responsibly.

Перевод пока недоступен