Transition State Theory-Inspired Neural Network for Estimating the Viscosity of Deep Eutectic Solvents

Area of Science:

• Green Chemistry
• Materials Science
• Computational Chemistry

Background:

• Accurate viscosity prediction for solvents, especially complex ones like deep eutectic solvents (DESs), is a significant challenge.
• The limited viscosity data for DESs hinders their industrial scale-up, leading to trial-and-error application development.
• Developing reliable predictive models is essential for advancing the use of DESs as sustainable solvents.

Purpose of the Study:

• To develop and validate a novel computational model for predicting the viscosity of deep eutectic solvents (DESs).
• To overcome the limitations of existing methods and facilitate the industrial implementation of DESs.
• To establish a highly accurate and reliable tool for DES viscosity prediction based on fundamental theories.

Main Methods:

• Development of a Transition State Theory-inspired Neural Network (TSTiNet) model.
• Utilizing multilayer perceptron (MLP) to calculate parameters for the Transition State Theory-inspired Equation (TSTiEq).
• Verification of the TSTiNet model using the most comprehensive DES viscosity dataset available.

Main Results:

• The TSTiNet model achieved high accuracy, with an average absolute relative deviation of 6.84% and an R-squared value of 0.9805 on the test set.
• The model demonstrated superior generalization ability compared to traditional machine learning methods.
• TSTiNet significantly reduced prediction deviations while adhering to thermodynamic constraints.

Conclusions:

• The TSTiNet model provides an accurate and reliable method for predicting DES viscosity.
• The model requires only the structural information of DESs, simplifying data input.
• This advancement is expected to accelerate the industrial adoption of deep eutectic solvents.