Machine Learning Optimization of Lignin Properties in Green Biorefineries

Abstract

Novel biorefineries could transform lignin, an abundant biopolymer, from side-stream waste to high-value-added byproducts at their site of production and with minimal experiments. Here, we report the optimization of the AquaSolv omni biorefinery for lignin using Bayesian optimization, a machine learning framework for sample-efficient and guided data collection. This tool allows us to relate the biorefinery conditions like hydrothermal pretreatment reaction severity and temperature with multiple experimental outputs, such as lignin structural features characterized using 2D nuclear magnetic resonance spectroscopy. By applying a Pareto front analysis to our models, we can find the processing conditions that simultaneously optimize the lignin yield and the amount of beta-O-4 linkages for the depolymerization of lignin into platform chemicals. Our study demonstrates the potential of machine learning to accelerate the development of sustainable chemical processing techniques for targeted applications and products.