Machine Learning for Green Solvents: Assessment, Selection and Substitution

Area of Science:

• Green Chemistry
• Computational Chemistry
• Environmental Science

Background:

• Growing demand for sustainable solvents due to strict environmental regulations.
• Existing Solvent Selection Guides (SSGs) are limited in scope and data availability.
• Identifying greener alternatives requires balancing complex criteria like sustainability, cost, and performance.

Purpose of the Study:

• To develop a data-driven pipeline for assessing solvent sustainability and identifying greener substitutes.
• To expand the pool of evaluated solvents beyond the limited scope of current guides.
• To create a comprehensive database of green solvent metrics.

Main Methods:

• Trained and evaluated Gaussian Process Regression (GPR) and GPT models on the GlaxoSmithKline Solvent Sustainability Guide (GSK SSG).
• Utilized GPR to predict greenness metrics for 10,189 solvents, creating the GreenSolventDB.
• Combined predicted metrics with Hansen solubility parameters to find suitable greener alternatives.

Main Results:

• GPR model slightly outperformed language-based GPT models in predicting solvent greenness.
• GreenSolventDB was established as the largest public database of green solvent metrics.
• Predicted alternatives for hazardous solvents like benzene and diethyl ether showed good alignment with known substitutes.

Conclusions:

• The developed framework effectively quantifies solvent sustainability and identifies viable greener substitutes.
• This approach significantly accelerates the discovery and adoption of environmentally friendly solvents.
• The GreenSolventDB provides a valuable resource for researchers and industry.