Machine learning and COSMO-RS integration for predicting anthocyanin extraction from berries using eutectic solvents.

Area of Science:

• Green Chemistry
• Computational Chemistry
• Biotechnology

Background:

• Developing efficient, sustainable extraction methods for bioactive compounds from diverse plant materials is crucial.
• Understanding solvent-solute interactions and biomass heterogeneity is key for optimizing extraction yields.
• Traditional extraction optimization relies heavily on extensive experimental screening.

Purpose of the Study:

• To develop a hybrid computational framework integrating COSMO-RS and machine learning (ML) for predicting anthocyanin yields.
• To enable rapid optimization of green extraction processes using deep eutectic solvents (DES).
• To minimize experimental efforts in identifying optimal extraction conditions for complex natural matrices.

Main Methods:

• A hybrid computational framework combining COSMO-RS and ML algorithms was developed.
• A dataset of 299 experimental points across 15 berry types was used for training and validation.
• Gradient Boosting ML algorithm demonstrated highest accuracy (R² = 0.92), utilizing COSMO-RS solvent descriptors, process variables, and ethanol-based anthocyanin content.

Main Results:

• The Gradient Boosting model accurately predicted anthocyanin yields across various berry matrices and DES.
• Model validation using independent literature data and new experimental results showed strong agreement between predicted and experimental values.
• The predictive model successfully integrated solvent properties, process parameters, and biomass characteristics.

Conclusions:

• The integrated COSMO-RS and ML approach provides a powerful tool for optimizing sustainable bioactive compound extraction.
• This computational strategy significantly reduces the need for extensive experimental screening.
• The framework facilitates efficient development of green extraction processes for complex plant-based materials.