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

What if we could predict how well a green solvent extracts anthocyanins from berries before even touching a pipette? A new hybrid model does just that—with striking accuracy.

Designing green extraction processes for bioactive compounds has long been a game of trial and error. But now, scientists have built a smarter path. They combined COSMO-RS, a quantum-chemistry tool that predicts how solvents behave, with machine learning to forecast anthocyanin yields from berries using deep eutectic solvents (DES).

The team trained seven machine learning algorithms on 299 experimental data points covering 15 different berry types. Gradient Boosting emerged as the star, achieving an impressive R² of 0.92. The model cleverly blends solvent descriptors from COSMO-RS, key process variables, and a single biomass descriptor: the anthocyanin content obtained with ethanol.

Validation was rigorous. The model was tested against independent literature data and new experiments with different DES and berry matrices. Predicted yields matched experimental results closely, proving its reliability.

This approach slashes the need for exhaustive experimental screening, paving the way for rapid, sustainable extraction optimization. It's a leap toward smarter, greener chemistry—where computation guides our hands.

Key Points

• Hybrid COSMO-RS + ML predicts anthocyanin yields from berries.
• Gradient Boosting achieved R² = 0.92 on 299 data points.
• Validated with literature and new experiments, cutting screening costs.