Study highlights importance of dedicated exits for vulnerable populations in building evacuations
image: In recent years, new strategies such as automatic control, data mining, transfer learning, hybrid model building and soft sensor construction have enhanced the adaptability of ML in designing and optimizing the complex fermentation system, which represent the future direction of this field.
Credit: The authors
Their overview highlights how ML can identify optimal fermentation conditions, improve strain performance, and extend to new strategies such as automated control, hybrid modeling, and soft sensors. This integration, the researchers suggest, could significantly advance industrial bioproduction in medicine, food, cosmetics, and bioenergy by lowering costs and accelerating efficiency.
Fermentation is one of the oldest biotechnologies, harnessing microorganisms to produce metabolites from renewable resources. Today, it underpins sustainable production of diverse bio-based products, aligning with global goals of carbon neutrality and green growth. Traditional fermentation optimization relies on empirical experiments and statistical modeling, but these methods struggle with the complexity of dynamic interactions between microbes and their environment. Strain development remains central, yet realizing the full potential of engineered strains requires precise optimization of temperature, pH, oxygen, nutrient inputs, and feeding strategies. With the rapid advances in synthetic biology producing new engineered strains, the demand for faster, more reliable optimization strategies has intensified. Machine learning, as a subset of artificial intelligence, offers a data-driven way to uncover hidden patterns, predict outcomes, and reduce the experimental workload.
A study (DOI:10.1016/j.bidere.2025.100002) published in BioDesign Research on 26 February 2025 by Xin-Qing Zhao’s team, Shanghai Jiao Tong University, demonstrates that machine learning can accurately simulate fermentation systems, predict optimal parameters, and extend into automated control, transfer learning, hybrid models, and soft sensor construction.
At the core of ML-based fermentation optimization is experimental design, where methods such as factorial design, Plackett-Burman, orthogonal, uniform, and Box-Behnken are used to systematically evaluate the effects of multiple variables. These designs generate structured datasets that are ideal for ML model training. Algorithms such as Random Forests, Support Vector Regression, and Backpropagation-Artificial Neural Networks have shown strong performance in capturing nonlinear relationships and predicting fermentation outcomes. Deep learning models further enhance accuracy and robustness, particularly when experimental data are limited or noisy. Beyond basic prediction, ML models are evaluated and tuned using statistical metrics like RMSE and R², while optimization algorithms such as grid search, genetic algorithms, and particle swarm optimization refine both model hyperparameters and fermentation parameters. Multi-objective optimization strategies allow simultaneous improvement of yield, rate, and substrate use efficiency, balancing economic and biological goals. Importantly, the review details emerging strategies that expand ML’s role. Automated control systems based on ANN and reinforcement learning can dynamically adjust feeding and oxygen supply, improving productivity while minimizing waste. Data mining—using tools such as natural language processing—leverages published data to predict strain performance, though heterogeneity remains a challenge. Transfer learning offers the possibility of applying models trained on one strain to another with fewer experiments, accelerating optimization for new engineered microbes. Hybrid models combine mechanistic fermentation kinetics with ML to exploit the strengths of both approaches, enabling more reliable predictions even with complex metabolic networks. Finally, ML-driven soft sensors provide indirect but real-time estimation of critical variables, reducing reliance on offline assays such as HPLC.
This review brings together the algorithms, workflows, and future applications of ML in fermentation engineering. By predicting optimal conditions, reducing experimental workload, and supporting advanced strategies like hybrid modeling and soft sensors, ML is reshaping the landscape of bioproduction. The authors emphasize that no single algorithm performs best in all contexts, making the thoughtful selection of experimental design, modeling methods, and optimization strategies essential. ML-based optimization will be central to unlocking the full potential of engineered strains, advancing industrial biotechnology, and supporting global sustainability goals.
###
References
DOI
Original Source URL
https://doi.org/10.1016/j.bidere.2025.100002
Funding information
We are thankful for the financial support from the State Key Research and Development Program of China (No. 2022YFE0108500), and National Natural Science Foundation of China (No. 21978168).
About BioDesign Research
BioDesign Research is dedicated to information exchange in the interdisciplinary field of biosystems design. Its unique mission is to pave the way towards the predictable de novo design and assessment of engineered or reengineered living organisms using rational or automated methods to address global challenges in health, agriculture, and the environment.