McSynC: in vivo automatic Model calibration of Synthetic Circuits components

合成生物学是一门新兴的工程学科，有一个雄心勃勃的目标：让科学家有能力将新功能编程到细胞中，就像他们对计算机所做的那样。然而，尽管社区蓬勃发展，并取得了显著的成功，为细胞编写“起作用的算法”仍然非常耗时。这在很大程度上是因为，我们用来组装这种“算法”的构件，即所谓的“部件”，很少像预期的那样运行，因为人们通常对它们的工作/动态了解得很少。数学模型特别适合于解决这个问题；在工程中，它们通常用于正式描述系统的行为，设计/模拟/筛选它们的性能，并节省时间只将最佳解决方案带到原型阶段(基于模型的系统工程)。尽管SynBio是一门工程学学科，但到目前为止，它对数学模型的使用有限，主要是因为推断生物模型传统上被认为是昂贵和/或困难的。如果英国“8大技术”之一的SynBio要实现英国合成生物学战略计划中设定的(生物)规模经济的期望，我们需要加快基因电路原型的制作：生物系统需要一种“基于模型的系统工程”方法；模型推理必须更简单、更快，并最终降低成本。为此，我建议将最优实验设计(OED)和显微镜/微流体相结合，开发一个自动化模型校准的网络物理平台，即识别模型中的参数。在给定感兴趣的部分和初始模型的情况下，该系统将在计算机中识别信息量最大的实验以改进参数估计；立即在体内运行此类实验；使用新的实验数据来更新模型并为新模型设计最优实验，反复迭代，直到达到稳健的估计。除了自动化模型校准，我提出的方法还有三个主要好处：它允许获得并公开分享可靠的模型：(A)更快--因为每个模型携带的信息更多，需要的实验更少；(B)经济高效--因为微流体大大减少了试剂的使用，自动化使人的干预变得不必要；(C)以一种可重复的方式--因为推断中的所有数据和步骤都被跟踪并立即公开。作为原理的证明，我们将使用这种方法来填补酵母SynBio中的一个空白：缺乏遗传振荡器。尽管在酿酒酵母中从头开始构建合成振荡器的努力失败了，但最近的一项研究提出了三种策略，将现有的“开关式”电路IRMA变成振荡器。这些干预措施中的每一种都需要用具有特定动态行为的新电路来取代现有电路的一部分。我们将利用我们的平台来寻找新的部分(pEGT2、pho和pANB1)，并引导基因电路的“重构”。总而言之，我们将：1.开发、部署和测试一种闭环方法，以自动推断遗传部件的数学模型；2.为先前提出的将IRMA转变为振荡器的三个部件中的每一个建立并表征一个库；3.在它们的模型的指导下，识别最好的候选部件，并使用它们来重构原始网络；4.测试新电路的振荡并对它们进行表征。

Synthetic Biology is an emerging engineering discipline with an ambitious goal: empowering scientists with the ability to programme new functions into cells, just like they would do with computers. Despite a booming community and notable successes, however, writing "functioning algorithms" for cells remains extremely time-consuming. This is mostly due to the fact that the building blocks we use to assemble such "algorithms", so-called "parts", rarely behave as expected as their working/dynamics are generally poorly understood. Mathematical models are uniquely suited to address this problem; in engineering, they are routinely used to formally describe systems' behaviour, design/simulate/screen them for performance, and save time bringing only the best solutions to the prototyping stage (Model-Based Systems Engineering). Despite being an engineering discipline, SynBio has so far made limited use of mathematical models, mostly because inferring biological models has been traditionally perceived as expensive and/or difficult. If SynBio, one of the UK's "8 Great Technologies", is to meet the expectations for a (bio)economy of scale set in the UK Synthetic Biology Strategic Plan we need to accelerate gene circuits prototyping: a "Model-Based Systems Engineering" approach is needed for biological systems; model inference must be simpler, faster and ultimately cheaper. To this aim, I propose to combine Optimal Experimental Design (OED) and microscopy/microfluidics to develop a cyber-physical platform that automates model calibration, i.e. the identification of parameters in a model. Given a part of interest and an initial model, this system will identify in silico the most informative experiment to refine parameter estimates; immediately run such experiment in vivo; use the new experimental data to update the model and design an optimal experiment for the new model, iterating until robust estimates are reached. Besides automating model calibration, the approach I propose has three main benefits: it allows to obtain, and publicly share, reliable models (a) faster -as fewer experiments are needed if each carries more information, (b) cost-effectively -as microfluidics drastically reduces reagents' use and automation renders human intervention unnecessary, (c) in a reproducible way -as all the data and the steps in the inference are tracked and immediately made publicly available. As a proof of principle, we will use this approach to fill a gap in yeast SynBio: the lack of a genetic oscillator. Despite the failures in building synthetic oscillators from scratch in S. cerevisiae, a recent study suggested three strategies to turn an existing "switch-like" circuit, IRMA, into an oscillator. Each of these interventions requires parts of the existing circuit to be replaced by new ones with a specific dynamic behaviour. We will use our platform to find the new parts (pEGT2, pHO and pANB1) and guide the gene circuit "refactoring". In summary, we will: 1. Develop, deploy and test a closed-loop method to automatically infer mathematical models of genetic parts;2. Build and characterise a library for each of the three parts previously proposed to turn IRMA into an oscillator; 3. Identify, guided by their models, the parts that are the best candidates and use them to refactor the original network;4. Test the new circuits for oscillations and characterise them.