Molecular machine learning in chemical process design

Molecular machine learning (ML) has recently demonstrated great potential in (i) predicting properties of pure components and their mixtures, and (ii) exploring the chemical space. We review state-of-the-art molecular ML models, such as graph neural networks and transformers, and discuss research directions for further advancements in chemical process engineering. This includes leveraging molecular ML at the process scale, for example, in process design and optimization formulations, which promises to accelerate the identification of novel molecules and processes. To this end, it will be essential to create design benchmarks and practically validate proposed candidates, possibly in collaboration with the chemical industry.