Data-Driven Bioprocess Optimization: Lipase Production by Aspergillus Niger ATCC 1004 in COCOA and Palm Oil Byproducts Based Solid State Fermentation

Abstract

This study aimed to optimize the production of lipase via solid-state fermentation (SSF) by Aspergillus niger ATCC 1004, using cocoa bean shell (CBS) and palm oil sludge (POILS). An artificial neural network (ANN) integrated with a particle swarm optimization (PSO) algorithm was employed. Fermentation time (days), initial moisture (%), and POILS (%) were evaluated using a Box-Behnken design. ANN performance was assessed based on the determination coefficient (R²) and mean squared error. The PSO tested two spaces: a restricted search region (RSR, -2 to + 2) and a broader search region (BSR, -3 to + 3). The 28-neuron ANN achieved R2 = 0.9998 (training), R2 = 0.97 (testing), R2 = 0.91 (validation), and R2 = 0.97 (overall). The predicted optimal lipase yields were 621.80 U g− 1 (broader region, BR) and 424.43 U g− 1 (restricted region, RR). Experimental validation yielded 322.50 ± 4.24 U g− 1 (BR) and 430.00 ± 3.31 U g− 1 (RR). The RR demonstrated better predictive accuracy, with experimental values reaching 98.7% accuracy, optimizing the response by 150% relative to the value predicted by the quadratic model. The optimal validated conditions for SSF were 4.4 days, 48% initial moisture, and 10% POILS. These results confirm the efficiency of PSO in optimizing ANN with limited datasets, reducing additional experiments and costs. POILS proved effective in stimulating lipase secretion, offering an alternative application in the bioprocess and paving the way for future research related to production expansion, and life cycle assessment of lipase production under the identified conditions.