Mohammad Suleiman Ibrahim, Owida Hamza Abu, Vasudevan Asokan, Ballal Suhas, Al-Hasnaawei Shaker, Ray Subhashree, Talniya Naveen Chandra, Sinha Aashna, Jain Vatsal, Abumalek Ahmad
Electronic Marketing and Social Media, Economic and Administrative Sciences, Zarqa University, Zarqa, Jordan.
INTI International University, Nilai, Negeri Sembilan, 71800, Malaysia.
Sci Rep. 2025 Aug 20;15(1):30569. doi: 10.1038/s41598-025-16150-x.
This research investigates the impact of bacterial growth on the pH of culture media, emphasizing its significance in microbiological and biotechnological applications. A range of sophisticated artificial intelligence methods, including One-Dimensional Convolutional Neural Network (1D-CNN), Artificial Neural Networks (ANN), Decision Tree (DT), Ensemble Learning (EL), Adaptive Boosting (AdaBoost), Random Forest (RF), and Least Squares Support Vector Machine (LSSVM), were utilized to model and predict pH variations with high accuracy. The Coupled Simulated Annealing (CSA) algorithm was employed to optimize the hyperparameters of these models, enhancing their predictive performance. A robust dataset comprising 379 experimental data points was compiled, of which 80% (303 points) were used for training and 20% (76 points) for testing. The study focuses on three bacterial strains including Pseudomonas pseudoalcaligenes CECT 5344, Pseudomonas putida KT2440, and Escherichia coli ATCC 25,922 cultured in Luria Bertani (LB) and M63 media, across varying initial pH levels, time intervals, and bacterial cell concentrations (OD600). Key input variables for the models included bacterial type, culture medium type, initial pH, time (hours), and bacterial cell concentration, all critical to pH dynamics. Sensitivity analysis using Monte Carlo simulations revealed bacterial cell concentration as the most influential factor, followed by time, culture medium type, initial pH, and bacterial type. The dataset was rigorously validated before training to ensure its suitability for predictive modeling. Evaluation of model performance demonstrated that the 1D-CNN model exhibited enhanced predictive precision, attaining the minimal RMSE and the maximum R² values and MAPE percentages in both training and testing phases. These findings underscore the efficacy of artificial intelligence techniques, particularly 1D-CNN, in precisely predicting pH changes in culture media due to bacterial growth. This methodology provides a reliable, cost-effective, and efficient alternative to traditional experimental approaches, enabling researchers to forecast pH behavior with greater confidence and reduced experimental effort.
本研究调查了细菌生长对培养基pH值的影响，强调了其在微生物学和生物技术应用中的重要性。一系列复杂的人工智能方法，包括一维卷积神经网络（1D-CNN）、人工神经网络（ANN）、决策树（DT）、集成学习（EL）、自适应提升（AdaBoost）、随机森林（RF）和最小二乘支持向量机（LSSVM），被用于高精度地建模和预测pH值变化。采用耦合模拟退火（CSA）算法优化这些模型的超参数，提高其预测性能。编制了一个包含379个实验数据点的可靠数据集，其中80%（303个点）用于训练，20%（76个点）用于测试。该研究聚焦于在不同初始pH水平、时间间隔和细菌细胞浓度（OD600）下，在Luria Bertani（LB）和M63培养基中培养的三种细菌菌株，包括假产碱假单胞菌CECT 5344、恶臭假单胞菌KT2440和大肠杆菌ATCC 25922。模型的关键输入变量包括细菌类型、培养基类型、初始pH值、时间（小时）和细菌细胞浓度，所有这些对pH动态都至关重要。使用蒙特卡罗模拟的敏感性分析表明，细菌细胞浓度是最具影响力的因素，其次是时间、培养基类型、初始pH值和细菌类型。在训练之前对数据集进行了严格验证，以确保其适用于预测建模。模型性能评估表明，1D-CNN模型在训练和测试阶段均表现出更高的预测精度，达到了最小RMSE以及最大R²值和MAPE百分比。这些发现强调了人工智能技术，特别是1D-CNN，在精确预测由于细菌生长导致的培养基pH变化方面的有效性。这种方法为传统实验方法提供了一种可靠、经济高效的替代方案，可以让研究人员更有信心地预测pH行为，并减少实验工作量。
