A Genetic Algorithms‐Based Neural Network Model to Monitor Gibberellic Acid GA3 Fermentation Process by Fusarium fujikuroi

• Published in: Biotechnology and bioengineering Vol. 122; no. 8; pp. 2111 - 2121
• Main Authors: Huang, Jia‐Cong, Wu, Jun‐Lin, Nie, Zhi‐Kui, Shi, Tian‐Qiong
• Format: Journal Article
• Language: English
• Published: United States Wiley Subscription Services, Inc 01.08.2025
• Subjects: Accumulation; Algorithms; Artificial neural networks; Biological activity; Carbon sources; Dissolved oxygen; Fermentation; Fusarium - genetics; Fusarium - growth & development; Fusarium - metabolism; Fusarium fujikuroi; genetic algorithm; Genetic Algorithms; Gibberellic acid; Gibberellins; Gibberellins - analysis; Gibberellins - metabolism; Glucose; Glucose - metabolism; Metabolites; neural network; Neural networks; Neural Networks, Computer; Optimization; Process control; Process controls; genetic algorithm; fermentation; Fusarium fujikuroi; neural network; gibberellic acid
• ISSN: 0006-3592; 1097-0290; 1097-0290
• DOI: 10.1002/bit.29022

ABSTRACT A genetic algorithm‐optimized neural network (ANN‐GA) was developed for real‐time monitoring of gibberellin (GA3) production during Fusarium fujikuroi fermentation. This model addresses the limitations of traditional off‐line detection methods, such as contamination risks and delayed feedback, by integrating six critical inputs—initial glucose concentration, fermentation time, temperature, pH, dissolved oxygen, and rotational speed—to predict glucose consumption and GA3 synthesis with an accuracy of 99.41%. During the implementation phase, by dynamically controlling the temperature (28°C–32°C) and pH, the biomass accumulation rate increased by 84% within 48 h, while the GA3 accumulation rate improved by 66.7% compared to constant‐temperature fermentation at 28°C. The ANN‐GA framework enables dynamic adjustment of glucose supply based on real‐time predictions, thereby optimizing carbon source utilization and enhancing process stability. This data‐driven approach effectively overcomes the drawbacks of costly sensors and labor‐intensive manual sampling, showcasing significant potential for industrial‐scale fermentation optimization. With its high accuracy and adaptability, the model holds substantial application value in advancing intelligent biological process control for secondary metabolite production.