Predictive design of crystallographic chiral separation.

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Machine Learning

Background:

• Efficient separation of chiral molecules is crucial for pharmaceuticals and materials science.
• Traditional methods rely on time-consuming trial-and-error processes.
• Developing predictive models for chiral resolution is a significant scientific challenge.

Purpose of the Study:

• To develop a machine learning-based approach for predicting resolving agents for chiral molecules.
• To improve the efficiency and reduce the cost of chiral resolution processes.
• To provide a publicly accessible dataset for advancing research in chiral separations.

Main Methods:

• Utilized a transformer-based neural network combined with a physics-based representation.
• Trained the model on a proprietary dataset of over 6000 chiral resolution experiments.
• Validated the model through retrospective testing and prospective experiments on unseen racemates.

Main Results:

• Achieved a four to six-fold improvement in hit rate compared to historical methods.
• Successfully resolved three out of six unseen racemates in a single experimental round.
• Demonstrated an 8-to-1 true positive to false negative ratio in prospective validation.

Conclusions:

• The developed machine learning approach significantly enhances the prediction of resolving agents for chiral molecules.
• This method offers a faster, more cost-effective alternative to traditional trial-and-error techniques.
• The release of the large-scale dataset and the predictive model will accelerate future research and development in chiral resolutions.