International Journal of Physics, Chemistry, Mathematics and Biology

Artificial intelligence (AI) has emerged as a transformative tool for predicting chemical reactions and estimating product yields, offering substantial benefits for sustainable and efficient chemical synthesis. This review synthesizes the theoretical foundations, datarepresentation schemes, and machine-learning paradigms that enable accurate reaction and yield prediction. Modern AI models—including graph neural networks, template-free architectures, and transformer-based sequence models—learn reactivity rules directly from large reaction datasets, overcoming several limitations of early expert-system and quantumchemistry-driven approaches. Data quality and curation remain central challenges, as public reaction databases often contain inconsistencies, structural biases, and incomplete yield information. Strategies such as standardized cleaning pipelines, benchmarked data splits, and uncertainty-quantification methods (e.g., ensembles, conformal prediction) improve model reliability and generalization. Recent advances demonstrate promising cross-domain transfer, enabling models trained on one reaction class to perform well on others. Incorporating reaction conditions, catalysts, and multimodal inputs further enhances predictive performance and selectivity estimation. Prospective experimental validation, transparent reporting standards, and open-science frameworks are essential to ensure reproducibility and trust in AI-assisted chemical design. Overall, AI-guided prediction of reaction outcomes and yields provides a powerful route toward accelerated discovery, greener synthesis pathways, and improved decision-making in chemical research and industry.