MFB: NSF-BSF: Data-Adaptive and Metamorphosis Machine Learning Architectures for Generative Protein Design of Metal Biosensors

堪萨斯大学的乔安娜·斯莱斯基博士与以色列海法大学的玛格丽塔·奥萨奇博士和雷切尔·科洛德尼博士合作，正在设计基于蛋白质的污染物生物传感器，可用于创造更清洁的水道。支撑传感器创造的技术将包括新颖的机器学习体系结构。这些新的ML体系结构可以用于开发生物传感器，用于更广泛的生物技术中的其他应用。这项研究将为来自堪萨斯州的本科生创造机会，学习自然如何控制金属结合，获得跨越生物学和计算机科学交叉的知识和技能，并参与国际合作。该项目涉及使用现有的和新开发的机器学习体系结构来生成用于生物传感器的新支架，特别是金属传感蛋白质。膜β桶是用于各种分析物生物传感器的理想平台，因为它们的形状、半径的多样性以及禁止桶内化学成分的能力。此外，由于这些蛋白质跨膜，分析物结合的行为会引起跨膜电导的可测量变化，这可用作生物传感器的输出。该项目有三个雄心勃勃的ML目标：1)使用较少的训练数据创建折叠特定的语言模型；2)利用基于斑块的变形、局部感知的迭代设计生成新的蛋白质序列；以及3)设计和测试候选蛋白质桶作为水污染物的生物传感器。这项研究追求一种限制人们对基于变压器的语言模型的关注的新方法，一种使用基于变形的生成器遍历蛋白质序列空间的新方法，更深入地理解膜β桶，新生成的生物传感支架，更深层次地理解是什么使一些蛋白质位点成为金属结合的理想条件，以及一种可以用于识别水中重金属污染物的蛋白质的设计方法。这一美国和以色列的合作项目得到了美国国家科学基金会和以色列双国科学基金会的支持。在美国国家科学基金会，该项目得到了化学部、化学、生物工程、环境和运输系统部、信息和智能系统部以及分子和细胞生物科学部的支持。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。

Dr. Joanna Slusky from the University of Kansas in collaboration with Dr. Margarita Osadchy and Dr. Rachel Kolodny from the University of Haifa, Israel, are designing protein-based biosensors for pollutants that could be used in efforts create cleaner waterways. The technology undergirding the creation of the sensors will include novel machine learning architectures. These new ML architectures could be used to develop biosensors for other applications in biotechnology more broadly. The research will create opportunities for undergraduates from Kansas to learn about how nature controls metal binding, to acquire knowledge and skills that span the intersection of biology and computer science, and to participate in international collaborations.This project involves the use of existing and newly developed machine learning architectures for the generation of new scaffolds for biosensors, particularly metal sensing proteins. Membrane β-barrels are ideal platforms for biosensors for a wide range of analytes because of their shape, variety of radii, and the ability to proscribe chemical moieties on the interior of the barrel. Moreover, because these proteins span membranes, the act of analyte binding causes a measurable change in conductance across the membrane, which can be used as biosensor output. This project thas three ambitious ML objectives: 1) create a fold-specific language model using less training data, 2) generate novel protein sequences with patch-based metamorphosis, locally-aware iterative design, and 3) design and test protein barrel candidates as biosensors for water contaminants. The research pursues a new way to restrict attention for transformer-based language models, a new way to traverse protein sequence space with a metamorphosis based generator, a deeper understanding of membrane β-barrels, newly generated scaffolds for biosensing, a deeper understanding of what makes some protein sites ideal for metal binding, and a design method for proteins that could be used to identify heavy metal contaminants in water. This collaborative US/Israel project is supported by the US National Science Foundation and the Israeli Binational Science Foundation. At the US National Science Foundation, the project is supported by the Division of Chemistry, the Division of Chemical, Bioengineering, Environmental and Transport Systems, the Division of Information and Intelligent Systems, and the Division of Molecular and Cellular Biosciences.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.