Machine learning advances molecular crystal design and crystallization development

Machine learning (ML) is increasingly being utilized to optimize the research paradigm and shorten the time from discovery to application of novel functional materials, pharmaceuticals, and fine chemicals. A recent review published in Engineering highlights the significant progress of ML in molecular crystal design and crystallization development.

The review, titled “Progress of Machine Learning in Molecular Crystal Design and Crystallization Development,” was authored by Shengzhe Jia, Yiming Ma, Yuechao Cao, Zhenguo Gao, Sohrab Rohani, Junbo Gong, and Jingkang Wang. The paper provides an in-depth look at how ML can support material and drug design, predictive modeling, and process optimization.

ML algorithms have shown great potential in various fields, including neuroscience, computer science, and chemistry. In the context of molecular crystal design, ML can accelerate the development of solvates, co-crystals, and colloidal nanocrystals, improving the efficiency of crystal design. The review notes that “ML can effectively screen different solvents and predict their solvate effects,” which is crucial for understanding the properties of drugs.

In terms of crystallographic structure prediction, traditional methods like solid-state density functional theory (DFT) are computationally expensive. ML-based algorithms, however, can simplify these complex calculations and reduce the time required for product development. The paper presents examples where ML models have successfully predicted the crystallographic structures of various molecules, such as oxalic acid and maleic hydrazide.

The review also highlights the role of ML in predicting crystal properties. It states that “ML can achieve high-throughput prediction; especially, it can map correlations between data when the physical mechanisms are unclear.” This capability is particularly useful in the design of crystalline materials, where ML models can forecast properties like plasticity, thermal conductivity, and configurational energies.

Co-crystal formation, a critical task in drug development, has also benefited from ML. The review mentions that ML algorithms can guide the selection of co-formers with simple descriptors, boosting prediction probabilities. This advancement can significantly reduce the experimental effort required in designing molecular crystals.

In the realm of crystallization development, ML models have demonstrated their potential in predicting crystallization behavior and optimizing process control. The review discusses how ML can predict drug solubilities, metastable zone widths, and particle agglomeration behaviors. It also highlights the use of ML-based online image processing for extracting information about crystal products and tracking crystallization in situ.

The authors emphasize the importance of integrating ML with experimental results to improve model accuracy. They suggest that future ML models should be constructed based on physical constraints and experimental data to enhance their predictive capabilities. The review concludes by noting that while ML has made significant strides in molecular crystal design and crystallization development, further research is needed to address challenges such as data quality, model interpretability, and scalability.

The review underscores the transformative impact of ML on chemical engineering and materials science, offering a glimpse into the future of intelligent and automated crystallization processes.

The paper “Progress of Machine Learning in Molecular Crystal Design and Crystallization Development,” is authored by Shengzhe Jia, Yiming Ma, Yuechao Cao, Zhenguo Gao, Sohrab Rohani, Junbo Gong, Jingkang Wang. Full text of the open access paper: https://doi.org/10.1016/j.eng.2025.03.036. For more information about Engineering, visit the website at https://www.sciencedirect.com/journal/engineering.