Statistical moments to infer microbial phenotypes in communities despite unquantified variables.

定量预测群落动态对于具有所需特性的工程微生物群落非常重要，例如降解废物或产生益生菌。对于具有充分理解的相互作用的群落，尽管可以构建机制模型，但它们对应于微生物表型的参数，如化学物质的释放或消耗速率，是出了名的难以测量的。例如，如果代谢产物被另一物种迅速消耗，则难以测量释放率；在单一栽培中测量的比率通常不能捕捉到社区环境中的表型。传统上，确定性模型是对数据进行拟合的，但由于信息不足，推断出的参数往往是不正确的或不是唯一的，这种情况由于未量化的变量（如快速消耗的化学品或酶中间体）而恶化。在这里，为了在未量化变量的情况下推断机制模型的参数，我将利用实验重复的统计信息开发SMIP“统计矩来推断参数”。我的初步工作表明，可以通过以描述变量的“统计矩”（例如均值、方差、协方差）的动态方程的形式施加附加约束来推断出基础真值。在本提案中，我将1)形式化SMIP，推导统计矩来推断未量化变量的参数；2)验证SMIP与硅社区的地面真相；3)在我将在实验中测量的两个微生物系统的时间序列中测试SMIP。我的研究不仅将帮助实验者克服在相关社区环境中量化微生物表型的挑战，而且还将促进预测合成生态学。这一行动将帮助我整合我的兴趣在理论和实验研究，发展必要的技能，一个成功的跨学科独立的职业生涯。

Quantitatively predicting community dynamics is important for engineering microbial communities with desired properties, e.g. degrading waste or producing probiotics. For communities with well-understood interactions, although mechanistic models can be constructed, their parameters which correspond to microbial phenotypes such as rates of release or consumption of chemicals, are notoriously difficult to measure. For example, release rates are difficult to measure if metabolites are rapidly consumed by another species; and rates measured in monocultures often do not capture phenotypes in the community setting. Traditionally, deterministic models are fitted to data, but the inferred parameters are often incorrect or not unique due to insufficient information to constrain them, a situation worsened by unquantified variables such as rapidly consumed chemicals or enzyme intermediaries. Here, to infer parameters of mechanistic models despite unquantified variables, I will develop SMIP "Statistical Moments to Infer Parameters" by taking advantage of the statistical information of experimental replicates. My preliminary work indicates that ground truth values can be inferred by imposing additional constraints in the form of equations that describe the dynamics of "statistical moments" (e.g. mean, variance, covariance) of variables. In this proposal, I will 1) formalize SMIP, deriving statistical moments to infer parameters despite unquantified variables; 2) validate SMIP against ground truth in silico communities; and 3) test SMIP in time series of two microbial systems that I will measure in experiments. My research will not only help experimentalists overcome the challenge of quantifying microbial phenotypes in the relevant community environment but also facilitate predictive synthetic ecology. This action will help me to integrate my interests in theoretical and experimental research, developing the skills necessary for a successful interdisciplinary independent career.