15a. Augmented Intelligence for Microscopy Informatics

The emergence of big data gathering experiments in microscopy has created new challenges in maintaining standards for scientific integrity and reproducibility. The diversity of data, available microscope modalities, and larger number of collaborators increase the complexity of scientific asset management for evolving experiments. Recently developed scientific asset management software addresses the problem of maintaining scientific integrity compliance with complex experiments, but it still requires significant labor overhead to review data and enforce scientific integrity compliance. A scientific asset management system enhanced with augmented intelligence capabilities is proposed. The system would benefit research institutions across the country by streamlining the administrative tasks required to reach compliance standards for scientific integrity. Deep-learning software architectures such as natural language processing and convolutional neural networks are used to identify correlations among text or images and have been successfully implemented in unsupervised and self-supervised tasks such as clustering, classification, and dynamic document generation. These matured and tested software architectures provide a researcher with augmented intelligence for microscopy informatics since the labor required to organize collected images, produce formal documents, and perform compliance evaluation can be automated by a server-side application. In Phase I the software architectures required to empower a user with augmented intelligence will be designed and implemented into an application that wraps around an existing scientific asset management software application. The proposed software architectures will be trained in a self- supervised manner on data and publications available online. The completed application prototype will then be tested on real microscopy data from collaborators at national labs. This project addresses the shortage of manpower available for parsing, organizing, annotating, and enforcing compliance standards on scientific data. It will benefit research laboratories across the United States by automating the bulk of the data administration tasks, resulting in redirecting more energy and resources towards pursuing scientific inquiries. This automation will allow microscopy research facilities across the nation to focus on training talent for performing experiments rather than on compliance standards that are subject to change and can vary in complexity relative to the experiments performed.