Transitions: Deep Learning Models for Microbial Image Analysis and Time-Series Predictions

Trustees Of Boston University, Boston

Investigators

Abstract

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117- 2). Deep learning is a powerful computational strategy for analyzing biological data. For example, models can be trained to identify the location of cells within an image, a task that has historically been cumbersome and required significant manual input. Although deep learning tools offer great potential for the analysis of biological data, taking advantage of state-of-the-art methods and avoiding potential pitfalls requires significant expertise. A major goal of this project is to provide the PI and research team applied training to take advantage of these empowering technologies. The project is divided into three periods. The first is a professional development period where the PI will learn state-of-the-art deep learning techniques through a combination of coursework, tutorials, and hands-on projects. The second part of the project focuses on applying these new deep learning methods to develop new tools for image analysis and time-series predictions. The third part spans all project years and involves education and outreach initiatives. These include partnering with the Engineering Biology Research Consortium to generate education modules on machine learning for researchers and educators in systems and synthetic biology. It will also introduce new curricular content that will be integrated into engineering coursework. In addition, the project provides training opportunities for undergraduate students. American Rescue Plan funding provides support for this investigator at a critical stage in her career. The primary sources of technical innovation for this project are the development of new tools for the analysis of time-lapse microscopy data and network inference algorithms. Underpinning these tools are deep learning models, including those based on convolutional and recurrent neural networks and transformers, which represent the current state-of-the-art for image processing and time-series predictions. The efforts will produce two classes of tools. In the first, the researchers will develop code to accurately segment, track, and determine cell fate within a series of time-lapse microscopy images. The second method uses time-series data to infer network connectivity. Deep learning models can handle complex relationships between signals such as time delays and feedback interactions, suggesting they may be a more accurate system identification tool than classical approaches. Overall, by taking advantage of new deep learning algorithms, the researchers will develop novel modeling approaches that are relevant to microbial image analysis and time-series predictions. 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.