MODELING EMERGENT BEHAVIORS IN SYSTEMS BIOLOGY: A BIOLOGICAL PHYSICS APPROACH

项目摘要 生物学上充满了令人惊叹的紧急行为的例子--这些行为源于，但不能被减少 即，构成所考虑的系统的各组成部分之间的相互作用。这些行为 从不同的细胞命运(如神经元、肌肉等)的出现，跨越了整个长度范围。 由于细胞内基因的相互作用，形成了复杂的生态群落 数千个物种之间的相互作用。我研究的首要目标是开发新的 概念、理论和计算工具，用于建模此类紧急的系统级行为 生物学。为了做到这一点，我们利用了以生物学和统计学为基础的跨学科方法 物理学，但这在很大程度上借鉴了机器学习、信息论和理论生态学。我们的工作 是统一和独特的，因为我们致力于将理论与大量的 生物数据现在正在通过实验产生。拟议研究的一个重要目标是找到 超越传统生物学分支学科和模型系统的共同概念和工具。 拟议的研究追求三个截然不同但在概念上相互关联的研究方向：(1) 确定控制微生物群落组装和发展的生态学原则 了解微生物群落的功能、多样性和稳定性的技术；(2)发展新的数学 以及用于对生化网络中的信息处理进行建模的计算工具，特别是基因 美国国立卫生研究院蜂窝身份的基础网络和控制集体行为的信令网络 (3)理解和开发新的可解释机器 系统和合成生物学的学习技术，特别注意独特的挑战 由生命系统提出的关于数据异构性、生物可解释性和潜在的 偏见。除了开发基于物理和机器学习的不同生物模型 现象，拟议的研究将产生一系列实用和重要的计算工具和 我们将公开提供的算法包括：(1)我们的“社区模拟器”Python包 为了模拟基于新型微生物消费者资源模型框架的微生物群落，我们 开发了新的机器学习和统计算法，用于分析微生物群落和 基因网络；以及(3)用于预测合成生物部分和 不同环境中的电路。这些计算工具将使科学家能够利用现代技术的力量 深度学习的理论、计算和进展，以解决与人类有关的基本问题 疾病。

Project Summary Biology is full of stunning examples of emergent behaviors – behaviors that arise from, but cannot be reduced to, the interactions of the constituent parts that make up the system under consideration. These behaviors span the full spectrum of length scales, from the emergence of distinct cell fates (e.g. neurons, muscle, etc.) due to the interactions of genes within cells, to the formation of complex ecological communities arising from the interactions of thousands of species. The overarching goal of my research is to develop new conceptual, theoretical, and computational tools to model such emergent, system-level behaviors in biology. To do so, we utilize an interdisciplinary approach that is grounded in Biological and Statistical Physics, but that draws heavily from Machine Learning, Information Theory, and Theoretical Ecology. Our work is unified and distinguished by our deep commitment to integrating theory with the vast amount of biological data now being generated by experiment. An important goal of the proposed research is to find common concepts and tools that transcend traditional biological sub-disciplines and model systems. The proposed research pursues three distinct but conceptually interrelated research directions: (1) identifying the ecological principles governing community assembly in microbial communities and developing techniques for understanding function, diversity, and stability in microbiomes; (2) developing new mathematical and computational tools for modeling information processing in biochemical networks, especially the gene networks underlying cellular identity and the signaling networks that control collective behavior in the NIH model organism Dictyostelium discoideum; and (3) understanding and developing new interpretable machine learning techniques for systems and synthetic biology, with special attention paid to the unique challenges posed by living systems with regards to data heterogeneity, biological interpretability, and potential sources of bias. In addition to developing physics-based and machine learning-inspired models for diverse biological phenomena, the proposed research will yield a series of practical and important computational tools and algorithms which we will make publically available including: (1) our “Community Simulator” Python package for simulating microbial communities based on the novel microbial consumer resource model framework we have developed; (2) new machine learning and statistical algorithms for analyzing microbial communities and gene networks; and (3) new computational tools for predicting the behavior of synthetic biological parts and circuits in diverse contexts. These computational tools will allow scientists to leverage the power of modern theory, computation, and advances in Deep Learning to tackle fundamental problems relevant to human disease.