Speaker
Description
Lignin is a structurally complex heteropolymer that has growing interest as a renewable feedstock for producing aromatic chemicals, biofuels, and bioplastics. One of the key obstacles in lignin valorisation is the compositional and structural variability of lignin obtained from different biomass sources, which makes it difficult to design efficient and generalizable processing strategies. Solvent fractionation is commonly used to narrow this heterogeneity, but choosing the right solvent is not straightforward, especially when lignin composition changes across feedstocks. In this work, we developed a computational approach to identify favourable and unfavourable organic solvents for lignin dissolution as a function of its monolignol composition. A library of 273 lignin polymer models was constructed using LigninBuilder, drawn from three biomass sources (birch, miscanthus, and spruce) to capture a wide and compositionally realistic range of syringyl, guaiacyl, and p-hydroxyphenyl monolignol ratios and interunit linkage distributions. Activity coefficients were calculated for each lignin–solvent combination across a large body of organic solvents using the conductor-like screening model based on segment activity coefficients (COSMO-SAC), which serves as a thermodynamically grounded metric for predicting dissolution tendency. We used this data, together with selected experimental solubilities, to train machine learning model (XGBoost) that allowed us to investigate how solvent properties, monolignol composition, molecular weight and interunit linkage distributions influence predicted lignin solubility. Leveraging the predictive power of this surrogate model alongside Kamlet–Taft solvent parameters, we mapped solubility behaviour across wide ranges of solvent polarity, basicity, and hydrogen bond donating ability, uncovering regions of solvent property space associated with favourable lignin dissolution regardless of feedstock origin. We also identified solvents that show selective dissolving behaviour depending on the monolignol composition of the lignin, suggesting their potential use in isolating fractions enriched in particular subunits. These outcomes provide guidelines for solvent selection that can be used to design fractionation processes tailored to lignin feedstocks of varying composition, as well as to identify solvents that promote lignin solubility regardless of monolignol ratios.
This work was supported from the LUABA24070 project of The Ministry of Education, Youth and Sports of the Czech Republic.
Keywords: Lignin solubility; Solvent selection; Machine learning; Monolignol composition