Multivariate regression and artificial neural network modelling of sugar yields from acid pretreatment and enzymatic hydrolysis of lignocellulosic biomass

Reducing sugar generation from lignocellulosic biomass (LCB) is closely linked with biomass characteristics, pretreatment and enzymatic hydrolysis conditions. In this study curated experimental data from literature was used to develop multivariate regression and artificial neural network (ANN) model considering nine predictors (i.e., cellulose, hemicellulose, lignin content, cellulose-lignin ratio, acid concentration, temperature, time, pretreatment severity, and enzyme concentration). Selected reduced polynomial model (R²: 0.891, Adj. R²: 0.849) suggests positive influence of acid and enzyme, while negative influence of treatment severity, temperature and time on reducing sugar generation. Genetic algorithm-optimized ANN model offered excellent fitness for LCB hydrolysis on training (R²: 0.997), validation (R²: 0.984), and test sets (R²: 0.967). Sensitivity analysis of the ANN predictors suggests lignin and to some extent hemicellulose contents can be inhibitory. Though polynomial models can have simple interpretation, use of optimized ANN offers better predictability in dataset with diverse biomass compositions.