Machine learning-driven optimization of pretreatment and enzymatic hydrolysis of sugarcane bagasse: Analytical insights for industrial scale-up

Abstract

The conversion of sugarcane bagasse (SCB) into fermentable sugars via pretreatment and enzymatic hydrolysis is a promising pathway for biomass valorization. However, the process’s complexity and variable optimization have limited its efficiency. This study introduces an orthogonal experimental design (OED) combined with machine learning (ML) to optimize NaOH-catalyzed Triton-X 100 pretreatment and enzymatic hydrolysis. The optimal pretreatment conditions identified through rule-based ML modeling (100 g/L solid loading, 45 g/L NaOH, 13.8 pH, 200 mL Triton-X 100, 175 °C, and 45 min) resulted in cellulose and hemicellulose recoveries of 88.5 % and 81.8 %, respectively, and a delignification of 92.3 %. The relative errors from experimental validation were 1.42 %, 0.56 %, and 3.55 % for these metrics, respectively. In the enzymatic hydrolysis (50 g/L substrate loading, 6 FPU/g enzyme loading, and 72 h hydrolysis), glucose and xylose yields reached 84.3 % and 63.3 %, with relative experimental validation errors of 1.11 % and 2.26 %, respectively. Key factors included time (26.2 % contribution to cellulose recovery), temperature (37.5 % to hemicellulose recovery), and solid loading (19.6 % to delignification). Substrate loading contributed 45.7 % to glucose and 37.8 % to xylose yields. This ML-optimized approach is projected to generate an additional US$321.45 million in profits by 2025, increasing to US$494.39 million by 2030, while reducing SCB waste by approximately 45 %. These findings highlight the potential of ML to enhance biomass conversion efficiency and accelerate the industrial adoption of bio-based sugar production systems.