MFB: Accelerating the Discovery of Novel Liposome Formations with Origins-of-Life Insights, Laboratory Automation, and Machine Learning

配方化学是一个关键但往往被忽视的领域，在制药、农用化学品、油漆和涂料、化妆品和家居产品等各种领域。现代设计的脂质双层结构是定义细胞的复杂的、多组分的混合物。在实现基本功能的情况下，它们可以简化到什么程度？了解如何构建具有特定功能的脂质结构可以促进关于生命起源和生物技术的基础知识。本项目开发的机器学习和自主研究方法对这个问题有直接的应用。设计的具有简化组成的脂质双层结构，如脂质体，对于新型生物药物、信使核糖核酸疫苗和农用化学品的药物输送具有重要意义。更具推测的是，创造人工的、功能最小的类细胞结构的能力可以与现有的无细胞生物化学系统相结合，产生将无细胞系统的工程优势与细胞系统的自我修复或自我支持相结合的新型合成生物系统。用简单的成分开发人造原细胞不仅可以让我们了解生命是如何进化的，还可以创造出经过工程改造的非生物生化系统。要做到这一点，我们必须用实验室自动化和机器学习方法克服人为偏见和组合爆炸。传统的化学进化方法是偏颇的，它以现存的生物体为基础，只考虑了相对有限的化学输入(10种反应物)来驯服组合的复杂性，从而对起始条件和反应物进行了“最佳猜测”。在这个项目中，研究人员将使用实验室自动化和机器学习指导的实验相结合的方式来获得数据集和统计基线，这些数据集和统计基线是测试探索和优化这些复杂、非理想混合物的算法所必需的。研究人员将开发自主配方化学的算法。与大多数机器学习数据集相比，实验化学数据存在噪声、有偏和小的问题，因此有必要在探索新的化学系统的同时利用现有的数据。研究人员将开发主动和元学习机器学习方法，以在处理新的优化问题时从现有的实验数据中学习，利用对比的元模型变化来推断相关变量。他们还将探索图形正则化矩阵分解方法，以直接从实验观察中学习低维表示。最后，他们将继续开发开源实验数据管理软件，以促进数据的重用和共享。在这个项目中，PI将通过举办公开挑战比赛，使用Kaggle等平台，并传播开放数据集，吸引更广泛的机器学习社区，目的是通过利用公民科学家池，将新的技术见解引入生命起源和生物物理学研究。这项研究将在中康涅狄格大学和福特汉姆大学两个仅面向本科生的化学系进行。该奖项将支持两所大学本科生的暑期和学年研究职位，以及两名博士后研究人员的研究。将博士后研究人员带入以本科生为重点的院系，使本科生接触到“科学家生活”的另一个阶段，特别是以“近乎同行”的形式，他可能比教授更有亲和力。它还使博士后研究人员有可能在非R1大学从事积极的研究工作。PIS将继续开发低成本、开源的机器人硬件和教学材料，将生命的起源和实验室自动化带入教学实验室，以帮助培训下一代化学家将自动化纳入他们的实验过程。该项目由化学部(CHE)、信息和智能系统部(IIS)、分子和细胞生物科学部(MCB)和物理部(PHY)共同支持。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Formulations chemistry is a crucial, but often overlooked area, in fields as diverse as pharmaceuticals, agricultural chemicals, paints and coatings, cosmetics, and household products. Modern designed lipid bilayer structures are complex, multicomponent blends that define the cell. How much can they be simplified and still achieve basic functionality? Understanding how to build lipid structures with specified functionality can advance both fundamental knowledge about origins of life and biotechnology. Machine learning and autonomous research methods developed in this project have direct applications to this problem. Designed lipid bilayer structures with simplified compositions, like liposomes, are important for drug delivery of novel biological pharmaceuticals, mRNA vaccines, and agrochemicals. More speculatively, the ability to create artificial, minimally functional cell-like structures could be combined with existing cell-free biochemistry systems to generate novel synthetic biological systems that combine the engineering advantages of cell-free systems with the ability to self-repair or self-support of cellular systems. Developing artificial protocells with simple components would not only inform our knowledge about how life evolved, but also enable the creation of engineered abiotic biochemical systems. To do this we must overcome the anthropogenic bias and combinatorial explosion with laboratory automation and machine-learning methods. Traditional approaches to chemical evolution have been biased by considering a “best guess” for starting conditions and reactants based on extant organisms and considered only a relatively limited numbers of chemical inputs ( 10 reactants) to tame combinatorial complexity. In this project the investigators will use a combination of laboratory automation and machine-learning-guided experimentation to obtain datasets and statistical baselines, needed to test algorithms for exploring and optimizing these complex, non-ideal mixtures. The investigators will develop algorithms for autonomous formulations chemistry. Experimental chemistry data is noisy, biased, and small compared to most machine learning datasets, and so it is necessary to both make use of existing data while also exploring new chemical systems. The investigators will develop active and meta- learning machine learning approaches to learn from existing experimental data when approaching new optimization problems, utilizing contrastive meta model changes to infer relevant variables. They will also explore graph regularized matrix factorization methods to learn low-dimensional representations directly from experimental observations. Finally, they will continue the development of open-source experimental data management software to facilitate data reuse and sharing. In this project the PIs will engage the broader machine-learning community by running open challenge competitions, using platforms like Kaggle, and disseminating open datasets, with the aim to bring new technical insights into origins-of-life and biophysics research, by drawing upon a pool of citizen scientists. This research will be conducted at two undergraduate-only chemistry departments at Central Connecticut University and Fordham University. This award will support summer and academic year research positions for undergraduate students at the two universities, as well as research of two postdoctoral researchers. Bringing postdoctoral researchers into undergraduate-focused departments exposes undergraduates to another phase of the “life of the scientist”, particularly in the form of a “near peer” who may be more relatable than a professor. It also exposes postdoctoral researchers to the possibility of active research careers at non-R1 universities. The PIs will continue the development of low-cost, open-source robotic hardware and pedagogical material that brings origins of life and laboratory automation into teaching labs, to help train the next generation of chemists to incorporate automation into their experimental process. This project is jointly supported by the Division of Chemistry (CHE), the Division of Information and Intelligent Systems (IIS), the Division of Molecular and Cellular Biosciences (MCB), and the Division of Physics (PHY) Physics.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.