Artificial intelligence in reaction prediction and chemical synthesis

Recent years have seen a sudden spurt in the use of artificial intelligence (AI) methods for computational reaction modeling and prediction. Given the diversity of the techniques, we believe it would be helpful to assess them using a broad conceptual framework within which the different approaches reside. Towards that goal, we categorize the different methods into symbolic AI, purely data-driven numeric AI, and hybrid AI methods. Symbolic AI-based approaches require the translation of a priori chemistry knowledge into clearly encoded rules and instructions. Purely data-driven numeric AI methods utilize recent advances in machine learning, generally without explicit incorporation of a priori domain knowledge. In between these two extremes, we have hybrid AI, which integrates domain knowledge with data-driven techniques. We review recent progress across these broad areas to highlight their benefits as well as limitations and provide a future outlook of this rapidly evolving field.

Automated summary

Recent years have witnessed a rapid increase in the use of artificial intelligence methods for computational reaction modeling and prediction. These methods can be broadly categorized into symbolic AI, purely data-driven numeric AI, and hybrid AI. Symbolic AI translates prior chemistry knowledge into rules, numeric AI utilizes machine learning without explicit domain knowledge, and hybrid AI integrates domain knowledge with data-driven techniques.