Unsupervised optimization of protein therapeutics using closed-loop in vitro synthesis, nanosensing, and deep-learning

使用闭环体外合成、纳米传感和深度学习对蛋白质疗法进行无监督优化

• 批准号: 10449253
• 负责人: Nigel F Reuel
• 金额: $ 36.32万
• 依托单位: IOWA STATE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-09-05 至 2025-06-30
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10449253
• 关键词: Affinity; Algorithms; Big Data; Binding; Binding Proteins; Biological Products; Biological Response Modifier Therapy; Biomanufacturing; Biosensor; Cell membrane; Cells; Clinic; Coupled; Custom; DNA biosynthesis; Data Set; Development; Devices; Disease; Engineering; Funding; Future; Generations; Genetic Template; In Vitro; Iowa; Label; Libraries; Machine Learning; Measurement; Membrane; Methods; Microfluidics; Optics; Patients; Pattern; Penetration; Phenotype; Property; Protein Biosynthesis; Protein Engineering; Proteins; Research; Small Business Innovation Research Grant; Speed; Stream; Surface; System; Techniques; Technology; Testing; Therapeutic Uses; Time; TimeLine; Training; Translating; United States National Institutes of Health; Universities; Vision; Work; base; bioelectronics; cost; deep learning; deep neural network; design; design-build-test; efficacious treatment; falls; improved; in silico; learning strategy; nanosensors; new technology; novel strategies; programs; prototype; radio frequency; sensor; side effect; synthetic biology; therapeutic protein; tool; treatment optimization; water quality; wound healing

Project Summary Overview of research: The Reuel Group at Iowa State University (founded Fall 2016) seeks to develop new materials, methods, and measurement devices for biomanufacturing, biotherapeutics, and biosensors. We have active work-streams in 1) optical nanosensors for protein binding, enzymatic activity, and cell membrane disruption, 2) scalable and reliable cell free protein synthesis (CFPS) methods for protein prototyping (extract and genetic template improvements), 3) microfluidics for droplet generation and measurement of CFPS products, interrogated by nanosensors, 4) algorithms for `big data' generated from nanosensors (machine learning and deep learning methods), 5) engineered endospores for time-delayed synthetic biology circuits, and 6) resonant radio frequency sensors for biomanufacturing, wound healing, and water quality. The overall vision of the research program is to simplify and improve the design and manufacturing of biological products (cells and proteins) for applications in therapies, advanced materials, and bio-electronics. Protein based therapies have demonstrated in clinic to be a potent tool in the treatment of many diseases. In recent years, the design, build, and test cycle to find therapies for new disease targets has improved dramatically using techniques such as surface display coupled to evolutionary selection. However, these mutagenic approaches have a few limitations, namely: 1) they require a suitable, naturally occurring sequence as a starting point, 2) they frequently optimize solely on a single desired feature, and 3) they operate as a `black box', meaning that generalizable design rules for in silico prediction of future products is not possible. It is the purpose of this MIRA for ESI research plan to design a closed-loop system that allows for unsupervised design and discovery of protein therapeutics that overcomes these limitations. Over the next five years we will build and integrate the system components which include enzymatic DNA synthesis coupled to cell free protein synthesis to rapidly prototype libraries of custom proteins in micro-droplet reactors. These proteins will then be characterized in the micro-droplets using optical nanosensors, to test for desired features such as stability, binding affinity, selectivity, hydrolytic activity, and/or membrane penetration. This will produce a large labeled data set (tying sequence to phenotypic properties) that can be used to train a deep learning neural network to self-determine sequence patterns for specific properties. Once the tuning coefficients of the network are found, the algorithm will then predict next best sequences which will be synthesized, tested, etc. such that the design loop progresses unsupervised until optimization criteria are met. This new approach will result in faster development of protein therapies that are optimized based on multiple criteria and not tied to existing, natural sequences. For patients this translates to more efficacious therapies with less side effects and a potential for reduced cost (due to shortened design timeline). At the end of the five-year project we will seek to translate this technology, via NIH SBIR funding, such that the new technology can make an impact on actual therapies.

项目摘要 研究概述：爱荷华州州立大学（成立于2016年秋季）的埃塞尔集团寻求开发新的 用于生物制造、生物治疗和生物传感器的材料、方法和测量装置。我们 在1）用于蛋白质结合、酶活性和细胞膜的光学纳米传感器方面具有活跃的工作流 破坏，2）可扩展和可靠的无细胞蛋白质合成（CFPS）方法，用于蛋白质原型（提取 和遗传模板改进），3）用于液滴生成和CFPS测量的微流体 产品，由纳米传感器询问，4）纳米传感器产生的“大数据”算法（机器 学习和深度学习方法），5）用于延时合成生物学电路的工程化内生孢子， 以及6）用于生物制造、伤口愈合和水质的谐振射频传感器。整体 该研究计划的愿景是简化和改进生物产品的设计和制造 （细胞和蛋白质）用于治疗、先进材料和生物电子学。基于蛋白质的 临床证明，这种疗法是治疗许多疾病的有效工具。近年来， 为新的疾病靶点寻找治疗方法的设计、建造和测试周期， 技术，如表面显示耦合到进化选择。然而，这些诱变方法 有一些限制，即：1）它们需要合适的、天然存在的序列作为起点，2） 它们经常只对一个所需的特征进行优化，3）它们作为一个“黑匣子”运行，这意味着 用于未来产品的计算机预测的可推广的设计规则是不可能的。这是我们的目的 MIRA for ESI研究计划设计一个允许无监督设计和发现的闭环系统 蛋白质疗法克服了这些局限性。在未来五年，我们将建立和整合 包括与无细胞蛋白质合成偶联的酶促DNA合成的系统组件， 在微滴反应器中快速建立定制蛋白质的原型库。这些蛋白质将被 其特征在于使用光学纳米传感器的微滴，以测试所需的特征，例如稳定性， 结合亲和力、选择性、水解活性和/或膜渗透性。这将产生一个大的标记 数据集（将序列与表型属性联系起来），可用于训练深度学习神经网络， 针对特定属性的自确定序列模式。一旦网络的调谐系数 如果找到了，则算法将预测下一个最佳序列，该下一个最佳序列将被合成、测试等，使得 设计循环在无人监督的情况下进行，直到满足优化标准。这种新方法将导致更快 开发基于多个标准优化的蛋白质疗法，而不依赖于现有的天然 序列的对于患者来说，这意味着更有效的治疗，副作用更少， 降低成本（由于缩短了设计时间轴）。在五年项目结束时，我们将努力翻译 这项技术，通过NIH SBIR基金，这样新技术就可以对实际治疗产生影响。