Statistical Inference from Multiscale Biological Data: theory, algorithms, applications

在过去的二十年里，从基因组学到流行病学，生命科学的不同领域都取得了巨大的实验性突破。由于现代高通量技术，现在可以在高空间和时间分辨率下对多个尺度的生物系统--从单个分子到整个种群--进行定量探测。除了增强我们对系统组成的基本知识外，这些数据还潜在地编码了大量信息，这些信息涉及支配系统演变的功能约束和限制其性能的物理约束，以及很难从低吞吐量数据中识别的组织级别、动态约束或设计原则。提取这些信息对于从设计具有所需功能的蛋白质到在流行病期间重建接触者的应用也是至关重要的。逆统计力学试图通过使用无序和随机系统物理学的方法从数据中推断生成模型(玻尔兹曼分布)来做到这一点。然而，生物学数据的特定特征，如严重的欠采样和异质性，限制了这些工具的有效性。辛巴德的目标是开发一类能够克服这些问题的统计推断技术。在SIMBAD，理论工作将提供概念和方法来解决四个紧迫问题(学习蛋白质序列图景、反向建模代谢网络、从流行病学数据推断接触网络和改进生存分析模型)，这反过来将指导该理论与每个领域的现有标准相结合。这一努力有望为基础研究开辟新的途径，以影响经济、技术和社会问题；SIMBAD所代表的高调跨学科专业知识确保实现可衡量和可实现的目标，使SIMBAD处于实现其目标的理想位置。

The last two decades have witnessed giant experimental breakthroughs in different areas of the life sciences, from genomics to epidemiology. Thanks to modern high-throughput techniques, biological systems across multiple scales -from single molecules up to entire populations- can now be probed quantitatively at high spatial and temporal resolutions. Besides enhancing our basic knowledge of a system's constituents, these data potentially encode a plethora of information about the functional constraints that govern its evolution and the physical constraints that limit its performance, as well as about levels of organization, dynamical constraints or design principles that would be hard to identify from low-throughput data. Extracting this information is also crucial for applications ranging from the design of proteins with a desired functionality to the reconstruction of contacts during an epidemics. Inverse statistical mechanics attempts to do it by inferring generative models (Boltzmann distributions) from data using methods from the physics of disordered and random systems. Specific characteristics of biological data however, like strong undersampling and heterogeneity, limit the effectiveness of these tools. SIMBAD aims at developing a class of statistical inference techniques capable of overcoming these issues. In SIMBAD, theoretical work will supply concepts and methods to address four pressing problems (learning protein sequence landscapes, inverse modeling metabolic networks, inferring contact networks from epidemiological data, and improving survival analysis models), which in turn will guide the theory towards integration with the existing standards of each field. This effort promises to open new pathways for basic research to impact economic, technological and societal issues; the high-profile cross-disciplinary expertise represented in SIMBAD ensures instead for measurable and achievable objectives, placing SIMBAD in an ideal position to achieve its goals.