Systems modelling of microbial communities using in vitro and computational approaches

了解和预测微生物群落对环境变化的反应对于基于微生物组的应用的发展至关重要。能够控制微生物组的组成和活性在许多领域都是有益的，包括生物技术、生物修复和动物/人类营养。先前的研究表明，通过了解微生物组的组成，我们可以预测其对特定干预的反应。体外微生物群落模型为高通量研究分子对微生物群落的影响提供了有用的系统。先前已经证明，体外微生物组允许在动物模型中复制的发现。这些方法还能够经济有效地提供大量信息，这些信息是用机器学习建模微生物群落所必需的。为了研究微生物组，科学家们对细菌群落的基因组进行了测序。这提供了大量的DNA序列，必须仔细解释。在大多数研究中，研究人员量化了细菌分类起源的丰度，并确定了细菌基因组中编码基因的代谢功能。然而，肠道微生物群中只有35%到45%的基因与实际功能有关。这往往限制了微生物组相关研究的分析和解释，并忽视了微生物群落的广泛功能和生态可能性。在这个研究项目中，我们的目标是为微生物群落的定向调节创建一个框架的基础。为了做到这一点，我们将解决两个关键问题，在我们可以故意调节微生物群以获得特定效果之前，这两个问题需要解决。首先，我们将设计一种新的方法来表示微生物组，这种方法允许有效地考虑所有基因和细菌物种，包括那些功能未知的基因和细菌物种。我们将使用机器学习来发现数据中有意义的模式，这些模式可能被当前基于细菌定量和基因功能分析的方法所忽视。其次，我们将使用机器学习来预测微生物组对特定条件的反应，并确定最佳干预顺序，以获得所需的微生物组效果。我们将使用体外培养粪便样品在三种微量矿物质的存在为模型，开发新的方法，使目标微生物组调节成为可能。使用机器学习来模拟微生物组对其化学环境的反应，将使人们更深入地了解微生物与其环境之间的相互作用。总的来说，我们的研究计划将提供方法来调节微生物组，这将适用于许多领域，包括生物技术和人类和动物的个性化营养。该项目还将为使用机器学习优化微生物群落的设计研究提供具体指导。

Understanding and predicting how microbial communities react to changes in their environment is critical for the development of microbiome-based applications. There are many fields where being able to control the composition and activity of microbiomes would be beneficial, including biotechnology, bioremediation and animal/human nutrition. Previous studies suggest that by knowing the composition of a microbiome, we could predict its response to specific interventions. In vitro models of microbial communities provide useful systems for the high-throughput study of the impact of molecules on microbiomes. It has been shown previously that in vitro microbiomes allowed discoveries that were replicated in animal models. These approaches are also able to cost-effectively provide a large quantity of information that is necessary to model microbial communities with machine learning. To study the microbiome, scientists sequence the genomes of bacterial communities. This provides large quantities of DNA sequences that must be carefully interpreted. In most studies, researchers quantify the abundance of the taxonomical origin of bacteria and determine the metabolic functions of the genes encoded in bacterial genomes. However, only 35% to 45% of genes from the gut microbiome can be associated with actual functions. This often limits the analysis and interpretation of microbiome-related studies and overlooks the extensive functional and ecological possibilities of microbial communities. In this research program, we aim to create the basis of a framework for the targeted modulation of microbial communities. To do so, we will address two critical aspects that need to be resolved before we can deliberately modulate microbiomes to get specific effects. First, we will devise new ways to represent microbiomes in a manner that allows to efficiently consider all the genes and bacterial species, including those with unknown functions. We will use machine learning to discover meaningful patterns in the data that may be overlooked using current methods based on bacteria quantification and gene function profiling. Second, we will use machine learning to predict the response of the microbiome to specific conditions and to determine the best sequence of interventions to obtain a desired microbiome effect. We will use in vitro culture of stool samples in presence of three trace minerals as a model to develop new methods to make possible targeted microbiome modulation. The use of machine learning to model the response of microbiomes to their chemical environment will permit a deeper understanding of the interplay between microorganisms and their environment. Overall, our research program will provide methods to modulate microbiomes that will be applicable to many fields, including biotechnology and personalized nutrition for both humans and animals. This project will also provide specific guidelines to design studies to optimize microbial communities using machine learning.