Microbial electrochemical sensors have been used to monitor water quality, with electroactive biofilms (EABs) serving as a core sensing element. However, since the bioelectric signals are incapable of recognizing different toxicants, the application of biosensors in complicated contamination is limited. With machine learning (ML) as a novel method to analyze bioelectric signals, we first quantified multiple toxicants with microbial electrochemical sensors. In this study, a batch of biosensors were shocked by mixed toxicants (MnCl₂, NaNO₂, and tetracycline hydrochloride (TCH)) at random concentrations. Regression ML models using different algorithms and datasets were developed and evaluated for prediction accuracy. The most accurate models for MnCl₂, NaNO₂, and TCH were trained with the algorithms of support vector machine, neural networks, and a generalized linear model. And the training set consisting of drop ratios at all time points showed the best accuracy for MnCl₂ and NaNO₂, while the most accurate model for TCH was trained with the drop ratio at 6 h. Here, we demonstrated that by integrating machine learning, a microbial electrochemical sensor is able to quantify multiple toxicants simultaneously, providing a fundamental of multiple-parameter biotoxicity detection for environmental monitoring.

中文翻译：

---

机器学习可以使用微生物电化学传感器对多种有毒物质进行量化

微生物电化学传感器已被用于监测水质，其中电活性生物膜 (EAB) 作为核心传感元件。然而，由于生物电信号无法识别不同的毒物，因此生物传感器在复杂污染中的应用受到限制。借助机器学习 (ML) 作为一种分析生物电信号的新方法，我们首先使用微生物电化学传感器量化了多种毒物。在这项研究中，一批生物传感器被随机浓度的混合毒物（MnCl ₂、NaNO ₂和盐酸四环素 (TCH)）冲击。开发并评估了使用不同算法和数据集的回归 ML 模型的预测准确性。最准确的 MnCl ₂ , NaNO模型₂ , TCH 使用支持向量机、神经网络和广义线性模型的算法进行训练。并且由所有时间点的下降率组成的训练集显示了 MnCl ₂和 NaNO ₂的最佳精度，而 TCH 最准确的模型是在 6 小时的下降率下训练的。在这里，我们证明了通过集成机器学习，微生物电化学传感器能够同时量化多种毒物，为环境监测提供多参数生物毒性检测的基础。