Building Predictive Coarse-Graining Schemes for Complex Microbial Ecosystems

微生物群落在全球气候、农业、食品安全和环境健康方面起着决定性的作用。这些系统高度复杂，往往由数百个相互作用的物种组成，这使得很难详细预测它们的行为。然而，众所周知，重要的生态系统属性(例如，一种养分的总生产量)有时可以在不跟踪每个物种的情况下预测，而是使用更简单、粗略的表示法。利用这种“粗粒度”可以改变我们预测和控制这些系统的能力，但这一现象仍然知之甚少：哪些属性预计是粗粒度的，以及在什么条件下，尚不清楚。该项目将利用理论物理学、统计学和学习理论的方法，建立生态系统粗粒化的系统理论，通过实验加以验证，并将其转化为向更广泛的研究界传播的实用软件工具。这项工作将显著提高我们对自然界中发现的复杂微生物生态系统的行为进行建模和预测的能力，例如那些负责全球养分循环、固氮或其他关键环境功能的微生物生态系统。在更技术性的层面上，该项目的目标包括：(1)建立量化生态系统粗粒化程度的理论和计算方法，确定特定群落一级可观测数据描述复杂性与其预测误差之间的帕累托折衷；(2)在实验室验证这一方法，将其应用于跨越海洋生态系统变化轴的数百个环境中的海洋细菌群落；(3)将该方法转化为可共享的软件，以确定成分微生物数据的特定可观测预测粗化。此外，该奖项将支持针对圣路易斯地区6-12年级学生的外展活动，强调对复杂系统可预测模型的追求如何将不同的科学领域(数学、物理、生物和医学，包括这项研究)结合在一起，并通过麻省理工学院暑期研究计划倡议增加未被充分代表的群体对研究的参与。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Microbial communities play a defining role in global climate, agriculture, food safety, and environmental health. These systems are highly complex, often composed of hundreds of interacting species, which makes it very difficult to predict their behavior in detail. However, it is known that important ecosystem properties (e.g., the overall production of a nutrient) can sometimes be predicted without tracking every species, instead using simpler, coarse representations. Leveraging such "coarse-grainability" could transform our ability to predict and control these systems, but the phenomenon remains poorly understood: which properties are expected to be coarse-grainable, and under what conditions, is not known. This project will draw on methods of theoretical physics, statistics and learning theory to develop a systematic theory of ecosystem coarse-grainability, validate it experimentally, and translate it into practical software tools disseminated to the broader research community. This work will significantly advance our ability to model and predict the behavior of microbial ecosystems of complexity as found in nature, such as those responsible for global nutrient cycling, nitrogen fixation or other key environmental functions. On a more technical level, the objectives of the project include: (1) Establishing the theoretical and computational methodology for quantifying ecosystem coarse-grainability and identifying the Pareto front of the tradeoff between description complexity and its prediction error for a given community-level observable; (2) Validating this approach in the laboratory, applying it to communities of marine bacteria assembled in hundreds of environments spanning the axes of variation relevant for ocean ecosystems; and finally, (3) Translating the methodology into shareable software for identifying observable-specific predictive coarsening of compositional microbial data. Additionally, this award will support outreach activities targeting students of grades 6-12 in the St Louis area, emphasizing how the pursuit of predictable models of complex systems unites diverse areas of science (math, physics, biology and medicine, including this research), and increase the participation of underrepresented groups in research through the MIT Summer Research Program initiative.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.