The potential of lignin as an abundant, underutilized biopolymer is increasingly being realized. A key challenge for the targeted production of lignins remains the poorly understood relation between lignin properties and its complex structure. Novel artificial intelligence (AI) methods could reveal such structure-function relationships but remain elusive in biomaterials research. Here, we present our AI study of the structure-function relationships in lignin carbohydrate complexes (LCC).
LCCs are extracted from birch wood combining the Aqua Solv Omni (AqSO) biorefinery process with AI-guided data acquisition by varying reactor temperature (T), reaction severity (P-factor) and liquid to solid (L/S) ratio [1]. Each LCC sample is characterized with 2D nuclear magnetic resonance (NMR) spectroscopy. A total of 95 NMR spectra are complemented with measurements of key lignin properties. As first examples, we focus on the antioxidant activity, glass transition temperature, molecular weight, surface tension and degradation metrics.
The analysis of 2D NMR spectra with AI is made feasibly through dimensional reduction of the 1024×1024 pixels of each NMR spectrum. We divide the spectra into rectangular regions, which are integrated to a single number and choose 70 of these integrated regions with the highest variance across samples.
To establish structure-function relationships, we first correlate these integrated regions of the NMR spectra with the corresponding property measurements. The employed random forest regression (RFR) model shows good predictive capabilities. Subsequently, we use RFR feature importance analysis to identify structural features that correlate with each property and provide a chemical interpretation of our findings, see Fig. 1. For instance, we find that a higher number of β-O-4 bonds leads to a lower surface tension in water indicating a more linear lignin structure. Our structure-inference approach is designed to be general and applicable to a wide range of materials and characterization data.
