We have introduced the use of multivariate NMR analysis in the development of accurate and robust prediction models, potentially arising from a correlation between soluble metabolite profiles and cell wall composition, for the determination of hemicellulose, cellulose and lignin contents in 8 species of greenhouse crop residues. The present paper demonstrates that discriminant buckets coming from a PLS-DA model in combination with linear models provide a useful and rapid tool for the determination of cell wall composition of these plant wastes. Regularized linear regression methods have also been applied to avoid overfitting, producing improved models specifically for lignin and cellulose determinations. The predictive models are also presented in a desktop application available at http://www2.ual.es/NMRMBC/solutions. To verify the rationality and reliability of the models, control experiments following generally accepted protocols have been performed and compared to our predicted values.
Keywords: Biomass; Cellulose; Greenhouse crop residues; Hemicellulose; Lignin; NMR; Predictive models.
Copyright © 2018 Elsevier Ltd. All rights reserved.