Programa de Doctorado en Genómica Integrativa, Vicerrectoría de Investigación, Universidad Mayor, Santiago, Chile.
Laboratorio de Redes Biológicas, Centro Científico y Tecnológico de Excelencia Ciencia & Vida, Fundación Ciencia & Vida, Santiago, Chile.
Microbiol Spectr. 2024 May 2;12(5):e0228723. doi: 10.1128/spectrum.02287-23. Epub 2024 Mar 20.
Understanding the interactions between microorganisms and their impact on bacterial behavior at the community level is a key research topic in microbiology. Different methods, relying on experimental or mathematical approaches based on the diverse properties of bacteria, are currently employed to study these interactions. Recently, the use of metabolic networks to understand the interactions between bacterial pairs has increased, highlighting the relevance of this approach in characterizing bacteria. In this study, we leverage the representation of bacteria through their metabolic networks to build a predictive model aimed at reducing the number of experimental assays required for designing bacterial consortia with specific behaviors. Our novel method for predicting cross-feeding or competition interactions between pairs of microorganisms utilizes metabolic network features. Machine learning classifiers are employed to determine the type of interaction from automatically reconstructed metabolic networks. Several algorithms were assessed and selected based on comprehensive testing and careful separation of manually compiled data sets obtained from literature sources. We used different classification algorithms, including K Nearest Neighbors, XGBoost, Support Vector Machine, and Random Forest, tested different parameter values, and implemented several data curation approaches to reduce the biological bias associated with our data set, ultimately achieving an accuracy of over 0.9. Our method holds substantial potential to advance the understanding of community behavior and contribute to the development of more effective approaches for consortia design.IMPORTANCEUnderstanding bacterial interactions at the community level is critical for microbiology, and leveraging metabolic networks presents an efficient and effective approach. The introduction of this novel method for predicting interactions through machine learning classifiers has the potential to advance the field by reducing the number of experimental assays required and contributing to the development of more effective bacterial consortia.
理解微生物之间的相互作用及其对群落水平细菌行为的影响是微生物学的一个关键研究课题。目前，有不同的方法，这些方法依赖于基于细菌不同特性的实验或数学方法，用于研究这些相互作用。最近，利用代谢网络来理解细菌对的相互作用的方法越来越多，这突出了这种方法在描述细菌方面的相关性。在这项研究中，我们利用细菌的代谢网络来构建一个预测模型，旨在减少设计具有特定行为的细菌群落所需的实验次数。我们预测微生物对之间的交叉喂养或竞争相互作用的新方法利用了代谢网络的特征。机器学习分类器用于从自动重建的代谢网络中确定相互作用的类型。我们评估和选择了几种算法，这些算法基于全面的测试和从文献来源手动编译的数据集的仔细分离。我们使用了不同的分类算法，包括 K 最近邻、XGBoost、支持向量机和随机森林，测试了不同的参数值，并实施了几种数据管理方法来减少与数据集相关的生物学偏差，最终实现了超过 0.9 的准确性。我们的方法具有很大的潜力，可以促进对群落行为的理解，并有助于开发更有效的群落设计方法。
重要性
理解群落水平的细菌相互作用对微生物学至关重要，而利用代谢网络则提供了一种高效有效的方法。通过机器学习分类器预测相互作用的这种新方法的引入，有可能通过减少所需的实验次数并有助于开发更有效的细菌群落来推动该领域的发展。
