Learning models of metabolism and gene expression from biological big data

背景 细胞代谢由维持重要细胞过程所需的数千个生化反应组成。细胞的代谢能力受到所表达的酶库的限制。计算模型，如基因组规模的代谢模型整合代谢与基因表达，从基因型预测表型。基因组规模的代谢模型通过将代谢反应制定为优化模型来预测细胞表型，该模型由生化目标（目标函数）驱动，同时受到物理化学性质，热力学和基因调控的约束。自90年代初以来，这些模型已成功应用于从可再生资源中生产有价值的化学品，以及生命科学和生物工程的知识进步。 研究计划 该研究计划的最终目标是开发计算机辅助设计（CAD）工具，以预测性地设计用于生产化学品，燃料和生物制药的基因工程细胞。一个先决条件是有精确的细胞代谢和基因表达模型。最近的建模进展允许E.大肠杆菌蛋白质表达预测高达85%的覆盖率。然而，对于其他生物技术上重要的生物体，如酵母或人类细胞，更高的生物复杂性和相对稀疏的机制知识使得实现如此广泛的模型范围具有挑战性。 该计划开发了新的CAD和建模工具，为E。大肠杆菌、酵母和人类细胞。为了解决这些生物体之间截然不同的生物复杂性和可用知识，开发了数据驱动和机械（知识驱动）建模方法。 1)对于人类细胞来说，机械知识是最重要的，新算法直接从“组学”数据（转录组学，蛋白质组学，通量组学）中学习优化模型。 2)对于酵母，构建了一个新的多尺度模型，该模型集成了代谢和基因表达。 3)大肠大肠杆菌，我们以前开发了先进的机械模型，CAD工具的开发和使用模型设计的菌株生产有价值的蛋白质。 开发的软件将免费分发给研究界。 对研究界和加拿大的好处 1)建模方法可用于在给定多种类型的组学数据的情况下提高对任何生物体的认识。 2)新的细胞设计工具可以应用于多个行业，包括生物制药制造，废物转化为化学品或燃料，以及多个平台生物（E。大肠杆菌、酵母、人细胞系）。 3)我将在计算系统生物学、基因组学和优化算法方面培养高素质的人才。

Background Cell metabolism consists of thousands of biochemical reactions needed to sustain vital cellular processes. The metabolic capabilities of a cell are constrained by the repertoire of enzymes expressed. Computational models such as genome-scale metabolic models integrate metabolism with gene expression to predict phenotype from genotype. Genome-scale metabolic models predict cell phenotype by formulating the metabolic response as an optimization model, driven by a biochemical goal (objective function) while being subject to constraints: physicochemical properties, thermodynamics, and gene regulation. These models have been applied successfully to produce valuable chemicals from renewable resources, and for knowledge advancement in the life sciences and bioengineering since the early 90s. Research program The ultimate goal of this research program is to develop computer-aided design (CAD) tools to predictively design genetically engineered cells for the production of chemicals, fuels, and biopharmaceuticals. A prerequisite is having accurate models of cell metabolism and gene expression. Recent modeling advances allow E. coli protein expression to be predicted with up to 85% coverage. However, for other biotechnologically important organisms like yeast or human cells, the higher biological complexity and relatively sparser mechanistic knowledge makes achieving such broad model scope challenging. This program develops new CAD and modeling tools for E. coli, yeast, and human cells. To address the vastly different biological complexity and available knowledge across these organisms, both data-driven and mechanistic (knowledge-driven) modeling approaches are developed. 1) For human cells, for which mechanistic knowledge is the sparsest, new algorithms learn optimization models directly from 'omics' data (transcriptomics, proteomics, fluxomics). 2) For yeast, a new multiscale model is constructed that integrates metabolism and gene expression. 3) For E. coli, for which we previously developed advanced mechanistic models, CAD tools are developed and used to produce valuable proteins using model-designed strains. The developed software will be distributed freely for the research community. Benefits to the research community and Canada 1) The modeling methods can be used to advance knowledge of any organism given omics data of multiple types. 2) The new cell design tools can be applied to multiple industries including biopharmaceutical manufacturing, and waste conversion to chemicals or fuels, and to multiple platform organisms (E. coli, yeast, human cell lines). 3) I will train highly qualified personnel in computational systems biology, genomics, and optimization algorithms.