Modeling Core

点击翻译按钮获取中文摘要

Modeling Core Project Summary: Many physical, biological, and social systems display sudden transitions between qualitatively different states. An example is the role of nutrient pollution on aquatic ecosystems: when phosphorus and nitrogen levels increase above a threshold value, a stream, river, or lake can transition from a low biomass/high diversity state to a high biomass/low diversity state. Systems that are history-dependent demonstrate hysteresis and follow so- called S-shaped bifurcation curves. We will approach the modeling of the development of pneumonia and its resolution in response to treatment using the conceptual framework of bifurcation theory. In the simplest case, with only two states, the model would distinguish a state with low bacterial load and high lung function from one with a high bacterial load and low lung function. The major challenge in this framework is to determine how to express the control parameter in terms of biological variables pertinent to pneumonia pathogenesis. We will pursue an agnostic modeling approach to the challenge of obtaining insight from these high-dimensional data. Because of the complexity of the problem, we will iteratively apply a variety of cutting-edge methods from systems biology, data science, dynamical systems, and ecology. By overcoming this challenge, we will achieve two aims. Aim 1. To identify biological variables (both host and pathogen) that will enable us to predict clinical outcomes in patients with Pseudomonas aeruginosa or Acinetobacter baumannii and other spp. pneumonia. Aim 2. To develop a set of hypotheses on the causal drivers of clinical outcomes that will be validated in subsequent human samples and tested using humanized mouse models. We will use systems biology methods to define low-dimensionality variables from the high-throughput, high-dimensionality data collected. We will then use machine learning, and dynamical systems methods — with a focus on methods that have demonstrated their mettle in ecological applications — to identify biomarkers for specific host/microbiome phenotypes and to predict the probability of different clinical outcomes for each phenotype. We will determine which biological variables most contribute to determining the classification by probing the sensitivities of different phenotypes to specific biological variables in order to generate mechanistic hypotheses that will then be tested experimentally with humanized mouse models and validated with human samples.