Predicting the Solubility of Lignin via Machine Learning.

Area of Science:

• Renewable Energy
• Polymer Science
• Computational Chemistry

Background:

• Lignin is a renewable resource with potential applications.
• Challenges include lignin's polydispersity and variable solubility.
• Predicting lignin solubility is crucial for its practical use.

Purpose of the Study:

• To develop a machine learning model for predicting lignin solubility.
• To understand how lignin molecular structure influences solubility.
• To guide the selection of green solvents for lignin processing.

Main Methods:

• Characterization of 100 lignins using GPC and HSQC NMR.
• Construction of lignin molecular structures.
• Machine learning model integrating structural and quantum chemical data.
• Solubility prediction using COSMOtherm software.
• SHAP analysis for feature importance.

Main Results:

• High accuracy achieved in predicting lignin solubility (R² values up to 0.987).
• Identified key molecular features affecting lignin solubility.
• Demonstrated the model's effectiveness across different solvents.
• Provided insights into structure-solubility relationships.

Conclusions:

• Machine learning accurately predicts lignin solubility based on molecular structure.
• Understanding structure-solubility relationships aids solvent selection.
• This approach facilitates the development of monodisperse lignin and sustainable solvent systems.