Predicting the Mixing Behavior of Aqueous Solutions Using a Machine Learning Framework.

Area of Science:

• * Chemistry and Materials Science: Focus on solution behavior and phase separation.
• * Computational Science: Application of machine learning for predictive modeling.

Background:

• * Traditional methods for determining aqueous solution phase separation are experimentally intensive and time-consuming.
• * Discovering new aqueous two-phase systems is crucial for various scientific and industrial applications.

Purpose of the Study:

• * To develop and validate a machine learning approach for predicting the miscibility of aqueous solutions.
• * To reduce the experimental burden in identifying potential aqueous two-phase systems.

Main Methods:

• * Experimental screening of 68 water-soluble compounds in 2278 unique two-component solutions to observe phase separation.
• * Training and validation of machine learning classifiers (artificial neural network, random forest, k-nearest neighbors, support vector classifier) using physicochemical properties.
• * Performance evaluation of classifiers based on accuracy, specificity, and sensitivity.

Main Results:

• * The random forest classifier demonstrated the highest performance with an average receiver operator characteristic area under the curve of 0.74.
• * Validated random forest model showed good predictive accuracy even when trained on limited data, with accuracy values around 0.70-0.74.
• * The model's ability to predict miscibility for compounds not present in the training data was assessed.

Conclusions:

• * Machine learning, particularly the random forest classifier, offers a promising approach to predict aqueous solution miscibility.
• * This predictive capability can significantly accelerate the discovery and design of aqueous two-phase systems.
• * The developed method has the potential to optimize screening experiments in chemical and industrial research.