Enhancing Lignin-Carbohydrate Complexes Production and Properties With Machine Learning

Abstract

Lignin-carbohydrate complexes (LCCs) present a unique opportunity for harnessing the synergy between lignin and carbohydrates for high-value product development. However, producing LCCs in high yields remains a significant challenge. In this study, we address this challenge with a novel approach for the targeted production of LCCs. We optimized the AquaSolv Omni (AqSO) biorefinery for the synthesis of LCCs with high carbohydrate content (up to 60/100 Ar) and high yields (up to 15 wt %) by employing machine learning (ML). Our method significantly improves the yield of LCCs compared to conventional procedures, such as ball milling and enzymatic hydrolysis. The ML approach was pivotal in tuning the biorefinery to achieve the best performance with a limited number of experimental trials. Specifically, we utilized Bayesian Optimization to iteratively gather data and examine the effects of key processing conditions–temperature, process severity, and liquid-to-solid ratio–on yield and carbohydrate content. Through Pareto front analysis, we identified optimal trade-offs between LCC yield and carbohydrate content, discovering extensive regions of processing conditions that produce LCCs with yields of 8–15 wt % and carbohydrate contents ranging from 10–40/100 Ar. To assess the potential of these LCCs for high-value applications, we measured their glass transition temperature (Tg), surface tension, and antioxidant activity. Notably, we found that LCCs with high carbohydrate content generally exhibit low Tg and surface tension. Our biorefinery concept, augmented by ML-guided optimization, represents a significant step toward scalable production of LCCs with tailored properties.