AI-driven optimization of sustainable solvent-based extraction: A deep learning approach for green sample preparation

Area of Science:

• Analytical Chemistry
• Green Chemistry
• Biotechnology

Background:

• Conventional plant biomolecule extraction methods are resource-intensive and environmentally demanding.
• A data-driven deep learning framework, Deep Green Optimization (DeepGO), was developed to address these limitations.
• DeepGO integrates sustainability indicators like solvent properties, energy demand, and toxicity into its model.

Purpose of the Study:

• To optimize the efficiency and sustainability of plant biomolecule extraction.
• To develop a predictive tool for analytical sample preparation.
• To reduce solvent consumption, energy demand, and waste generation in chemical practices.

Main Methods:

• A hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) deep learning model was employed.
• The model was trained on over five thousand experimental data points.
• Decision-making algorithms were utilized to identify optimal extraction parameters.

Main Results:

• The DeepGO framework accurately predicts extraction yields, phytochemical content, and bioactivities.
• Moderately toxic and polar solvents like n-butanol and ethyl acetate showed the highest extraction efficiencies (R²=0.87).
• Solvents of extreme polarity, hexane and water, demonstrated limited compatibility with conventional methods (0.44

Conclusions:

• DeepGO serves as a valuable predictive tool for enhancing plant-based extractions.
• The framework significantly reduces solvent usage, energy consumption, and waste.
• This approach promotes sustainable chemical practices in biomolecule extraction.