CISE-MSI: DP: III: Training and Partnership in Data Science for Advancing Research in Biomolecular Detection

检测和识别蛋白质、病毒、碳水化合物和DNA等生物分子对于医学诊断、法医分析、水质监测、食品安全、药物输送和材料设计等不同学科至关重要。从不同的光谱技术测量的数据提供了对这些生物分子的有价值的见解。分子化合物--无论是设计的还是天然的--是由多种不同的生物分子组成的，研究具有挑战性，包括分析它们错综复杂的数据。为此，数据科学结合了统计和计算方法，不仅可以分析大型复杂数据，还可以从数据中产生新的见解。该项目展示了数据科学和机器学习方法与选定生物分子的光谱测量相结合的方法。总体目标是开发一个自动数据驱动的模型，用于高效检测和准确识别生物分子，可用于加速发现和设计新技术。该项目提供了一系列机会，以培训在STEM领域历史上代表性不足的学生在数据科学、生物化学和物理之间的跨学科研究中。该项目侧重于从碳水化合物生物分子开始建立初始光谱数据库，并设计机器学习模型，以这些生物分子的多模式和异质光谱数据进行训练和测试。它有两个目标：1)展示可解释和可靠的物理信息机器学习模型，可以提高生物分子检测的效率和准确性；2)在处理混合或稀疏数据时，为基于知识的机器学习中的基本问题做出贡献，在有限的数据库中增加一维数据，并理解噪声数据与训练和测试精度之间的复杂关系。该项目建立在两所HBCU(特拉华州立大学和维尔京群岛大学)与研究机构特拉华大学之间的协作努力基础上。在领域指导的数据生命科学框架下，参与的学生将接触到整个数据科学方法周期，包括收集数据、对数据进行预处理和分析、设计机器学习模型、培训和测试模型、验证模型以及提供有关数据和模型的知情反馈。一旦得到证明，该方法的范围将扩展到其他生物分子，如蛋白质和复杂系统。该项目由MSI和既定的激励竞争性研究计划(EPSCoR)共同资助。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。

Detecting and identifying biomolecules such as proteins, viruses, carbohydrates, and DNAs are critical to diverse disciplines such as medical diagnosis, forensic analysis, water quality monitoring, food safety, drug delivery, and materials design. Data measured from diverse spectroscopic techniques provide valuable insights into these biomolecules. Molecular compounds -designed or natural- are made of multiple and diverse biomolecules, and are challenging to investigate, including analysis of their convoluted data. For this purpose, data science, which combine statistical and computational methods, can be integrated to not only analyze large complex data but also generate new insights from the data. This project demonstrates the integration of data science and machine learning approaches with spectroscopic measurements on selected biomolecules. The overarching goal is to develop an automatic data-driven model for efficient detection and accurate identification of biomolecules, which can be used for accelerating discovery and designing novel technologies. The project opens a range of opportunities to train students from groups historically underrepresented in STEM fields in interdisciplinary research at the interface between data science, biochemistry, and physics.The project focuses on building initially-spectral databases starting with carbohydrate biomolecules and on designing machine learning models to be trained and tested with multimodal and heterogeneous spectral data of these biomolecules. It has dual objectives: 1) to demonstrate interpretable and trustworthy physics-informed machine-learning models that can improve efficiency and accuracy of biomolecular detection; and 2) to contribute to the foundational questions in knowledge-based machine learning when handling mixed or sparse data, augmenting one-dimensional data in limited databases, and understanding the relational intricacies between noisy data and accuracies of training and testing. This project is built on collaborative efforts between two HBCUs (Delaware State University and the University of Virgin Islands) and the University of Delaware, a research institution. Under a domain-guided data life science framework, the participating students are exposed to a full cycle of data science approaches including collecting data, preprocessing and analyzing the data, designing machine learning models, training and testing the models, validating the models, and providing informed feedbacks about the data and the models. Once demonstrated, the scope of the approach will be extended to cover other biomolecules such as proteins and complex systems. This project is jointly funded by MSI and the Established Program to Stimulate Competitive Research (EPSCoR).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.