Predicting Crystalline Material Properties with AI: Bridging Molecular to Particle Scales

Abstract

Predicting crystalline material properties using artificial intelligence (AI) has seen significant advancement in recent years. This review aims to provide an in-depth overview of AI-based approaches for accurate and rapid prediction of chemical properties. First, various principles and models of machine learning (ML) are critically summarized, highlighting the strengths and weaknesses of different algorithms to assist researchers in selecting the most appropriate model for chemical properties prediction. Furthermore, the applications of AI in predicting the properties of metal−organic frameworks (MOFs), biomolecules, and energetic materials using molecular-scale descriptors are discussed, alongside its ability to predict the properties of lithium batteries and crystalline products at the particle scale. Compared to traditional methods (density functional theory calculations and molecular simulations) for discovering various properties, the predicted results reveal that AI-based algorithms exhibit superior capabilities, particularly in terms of accuracy, efficiency, and reduced computational costs. Lastly, challenges and opportunities in algorithm optimization, database enhancement, and feature descriptor exploration within this field are discussed.