Mixture design and machine learning-based optimization of fermentable sugar recovery from mixed agricultural waste for bioethanol production

The global interest in biofuel and value-added product generation from waste agricultural biomass is significant, however, current approaches often rely on single carbon sources, which are often constrained by seasonal substrate availability and low sugar yield. Combining different starch-based feedstocks can address these challenges, yet there is a dearth of reports on mixed starch-based biomass potential and the mixture design strategies required for their optimal use. Thus, with a focus on bioethanol production, this study models fermentable sugar (FS) release from thermochemically pretreated mixed starch-based agricultural wastes (MSBAW). The modelling was achieved through an optimized mixture design response surface methodology (MDRSM). Furthermore, regularized linear machine-learning model (ridge regression), linear regression and Artificial Neural Network (ANN) were trained to predict fermentable sugar release from the percentages of cellulose, hemicellulose, lignin, and starch units across the 14 MDRSM-designed mixtures; their performances were evaluated using leave-one-out cross-validation (LOOCV). Mixture integrity was confirmed by comparing theoretical vs. measured bioenergy compositions. Also, the role of starch, hemicellulose, cellulose, and lignin composition in FS recovery was assessed using sensitivity analysis. MDRSM and the regularized linear machine learning model gave the best correlation coefficients (R 2 ) of 0.94 and 0.67 respectively. Additionally, the sensitivity analysis based on the ridge model, showed compositional units with negative influences on FS release were starch and lignin while cellulose unit favours FS recovery within the observed mixture space. The validated MDRSM model gave an optimized FS concentration of 11.51 g/L. Subsequent fermentation of the pretreated and saccharified MSBAW hydrolysate gave maximum bioethanol concentration of 23.52 g/L using Saccharomyces cerevisiae . The study provides useful insight into the use of mixed starch-based agricultural wastes for FS recovery towards MSBAW-based bioethanol production. • First integrative modelling on mixed starch-based agricultural waste valorization. • Starch, cellulose, hemicellulose, and lignin units influence fermentable sugar recovery. • High fermentable sugar (11.86 g/L) was obtained with mixed feedstock pretreatment. • Mixed starch-based feedstock hydrolysate gave bioethanol concentration of 23.52 g/L.