结合实验与人工智能筛选牛至和薄荷精油中靶向白色念珠菌的抗真菌化合物及其与制霉菌素的协同效应

本研究创新性地融合人工智能（AI）与实验验证，系统评估了牛至（Origanum vulgare）和薄荷（Mentha pulegium）精油及其与制霉菌素（nystatin）联用对白色念珠菌（Candida albicans）的抗真菌（MIC）和抗生物膜活性。通过机器学习（如AdaBoost、Random Forest）预测关键活性成分（如百里香酚thymol、胡薄荷酮pulegone），并经实验证实其协同效应与模型可靠性，为抗耐药真菌感染提供了数据驱动的研发新策略。

　　

Highlight the role of ML predictions in improving the efficiency of the experimental plan and in reinforcing the biological relevance of the conclusions.

强调机器学习（ML）预测在提升实验方案效率及强化结论生物学相关性中的作用。

Machine learning models

机器学习模型

Three machine learning algorithms were used to predict the antifungal activity and antibiofilm properties of essential oils: Linear Regression offers a straightforward and interpretable approach, ideal for modeling linear relationships. AdaBoost was selected for its ability to improve accuracy by focusing on correcting specific errors, making it particularly effective for complex datasets. Random Forest captures intricate relationships and is robust against overfitting.

本研究采用三种机器学习算法预测精油的抗真菌活性与抗生物膜特性：线性回归（Linear Regression）提供直观可解释的线性关系建模；AdaBoost通过修正特定误差提升精度，适用于复杂数据集；随机森林（Random Forest）能捕捉复杂相互作用且抗过拟合性强。

Performance analysis of proposed machine learning models

机器学习模型性能分析

Figure 1 presents the concentrations predicted by three machine learning models: AdaBoost, Random Forest, and Linear Regression, compared to the experimental reference values for the chemical compounds O. vulgare and M. pulegium. These graphs illustrate the performance of each model in terms of prediction accuracy for the essential oils’ concentrations from both plants. The results show that the AdaBoost and Random Forest models yield predictions that closely align with the experimental values, while Linear Regression exhibits larger deviations, particularly for complex compositional profiles.

图1展示了AdaBoost、随机森林和线性回归三种模型对牛至与薄荷化学成分的浓度预测值与实验参考值的对比。图表显示，AdaBoost与随机森林的预测结果与实验值高度吻合，而线性回归模型偏差较大，尤其在处理复杂成分谱时表现显著不足。

Discussion

讨论

From Figure 1, a comparative analysis of the concentrations predicted by the AdaBoost, Random Forest, and Linear Regression models, in relation to the experimental values of the chemical compounds in O. vulgare and M. pulegium, reveals a spectrum of performance characteristics. The AdaBoost and Random Forest models exhibit high predictive accuracy, closely aligning with the experimental data, demonstrating their capacity to capture the intricate, non-linear relationships inherent in the essential oil compositions. In contrast, Linear Regression, while computationally efficient, struggles with the complexity of the data, resulting in less precise predictions. This underscores the superiority of ensemble methods like AdaBoost and Random Forest for modeling biologically complex systems such as essential oil interactions with microbial targets.

基于图1对三种模型预测浓度与实验值的比较分析可知：AdaBoost和随机森林模型展现出高预测精度，能有效捕捉精油成分中固有的复杂非线性关系；而线性回归虽计算高效，却难以处理数据复杂性，预测精度较低。这凸显了集成方法（如AdaBoost和随机森林）在模拟生物复杂系统（如精油与微生物靶点互作）中的显著优势。

Conclusion

结论

This study demonstrates the significant antifungal and antibiofilm effects of the essential oils from Origanum vulgare and Mentha pulegium, driven by their oxygenated monoterpene content, which suggests their potential for treating and preventing candidiasis. Notably, O. vulgare essential oil exhibits strong efficacy in inhibiting biofilms in vitro, with a synergistic interaction with nystatin against C. albicans. A comparative analysis of machine learning models, including AdaBoost, Random Forest, and Linear Regression, highlights the superior predictive performance of ensemble methods, supporting their utility in accelerating the discovery of novel antifungal agents and optimizing combination therapies.

本研究证实牛至与薄荷精油因其含氧单萜成分具有显著抗真菌与抗生物膜效应，有望用于念珠菌病的防治。尤其牛至精油在体外抑制生物膜效果突出，且与制霉菌素协同抗白色念珠菌。机器学习模型对比表明集成方法（如AdaBoost、随机森林）预测性能卓越，为加速抗真菌药物发现及优化联合疗法提供了有力工具。