Predicting the formation of NADES using a transformer-based model.

Area of Science:

• Green Chemistry
• Computational Chemistry
• Materials Science

Background:

• Natural deep eutectic solvents (NADES) offer sustainable alternatives to traditional organic solvents in various industries.
• Current NADES development relies heavily on empirical methods, limiting the discovery of novel mixtures.
• A data-driven approach is needed to overcome the limitations of traditional NADES discovery.

Purpose of the Study:

• To develop a machine learning model for predicting the formation of stable NADES.
• To accelerate the discovery and design of new NADES with desired properties.
• To reduce the reliance on empirical methods in NADES development.

Main Methods:

• A transformer-based neural network was employed, leveraging natural language processing techniques.
• The model was pre-trained on general chemical data (SMILES) and fine-tuned for binary classification of stable NADES formation.
• The approach adapted language learning strategies for efficient pattern recognition with limited data and computational resources.

Main Results:

• The algorithm successfully predicted 337 new stable eutectic mixtures from a natural compound database.
• The model demonstrated the ability to predict components and molar ratios for NADES incorporating novel molecules.
• Validation included previously reported NADES and the development of new ibuprofen-containing solvents.

Conclusions:

• The proposed machine learning strategy significantly enhances the prediction and discovery of NADES.
• This approach has the potential to revolutionize NADES screening and streamline the use of bioactive compounds in liquid formulations.
• The method offers a powerful tool for advancing green chemistry and the pharmaceutical industry.