Forrest Sheng Bao - US grants

Over the past decade, systems and synthetic biology approaches provided novel mechanism to enhance the production of diverse chemicals and biofuels from renewable resources in laboratory settings. However, it is still rare for synthetically modified strains to meet the production requirement for commercialization. Strain development falls into the tedious and costly design-build-test-learn cycle because existing modeling approaches failed to capture the complicated metabolic responses in such engineered cells. This proposal will explore an alternate, data-driven approach that has the potential to predict the productivity of synthetic organisms by leveraging the vast array of microbial cell factory publications. Using Artificial Intelligence approaches such as Machine Learning and Knowledge Representation, one can abstract "previous lessons'' hidden in published data to facilitate a priori estimations of the metabolic output by engineered hosts given a set of specific genetic instructions and fermentation growth conditions. The resulting platform can assist current constraint-based models to design the most effective strategies for producing value-added chemicals. On the educational front, this proposal will offer educational and research training opportunities in synthetic biology, computer programming, and artificial intelligence for graduate students to provide them with a non-conventional career pathway.

Synthetic biology relies on extensive genetic modification and pathway engineering, which often result in unexpected physiological changes or metabolic shifts that reduce the productivity and stability of the hosts. The investigators conceived of a creative, multidisciplinary approach that relies on artificial intelligence-inspired methods for predicting the performance of two distinct unicellular cell factories (Escherichia coli and Saccharomyces cerevisiae). These platforms can be used to quantify the factors that govern microbial productivity (yield, titer, and growth rate), including the type and availability of metabolic precursors; the elements that constitute a biosynthetic pathway; fermentation conditions; and the specific genetic modification to optimize the system. By extracting and classifying information derived from referenced publications within the last 20 years, one can construct a ''knowledge base'' containing sufficient samples of bio-production assemblies. This information will then inform the building of cellular factories using supervised machine learning and non-monotonic logic programming to estimate the productivity of hosts. The data-driven platform will also be integrated into genome scale models to project physiological changes of specific mutant strains. This novel approach will reduce the need for costly design-build-test bench work. Key outcomes from this project include: (1) a database to standardize synthetic biology studies, (2) machine learning models to recognize lessons and patterns hidden in published data, and (3) integration of machine learning with flux balance models, leading to the design of strains with high chances of success in industry settings.

Many research domains, such as high-energy physics, climate science, astrophysics, combustion science, and computational biology, need to process large amounts of data. Such domains are heavily relying on the capabilities of high performance computing (HPC) systems to manage and efficiently process massive amounts of data. Consequently, applications in the aforementioned research domains require highly optimized performance on the HPC storage systems that store, manage, and manipulate data. This project aims to utilize deep reinforcement learning methods to fine-tune the HPC storage system for optimized performance.

This research explores the feasibility of leveraging deep reinforcement learning to optimize HPC storage systems by: (a) Creating a deep learning based HPC storage stack model; (b) Remodeling existing HPC storage stack to support automated configuration and tuning; (c) Collecting training datasets and training the storage stack model; and (d) utilizing the model as a responsive and playable virtual environment to learn the best policy to tune parameters.

As a collaborative project, this research aims to advance the domain knowledge of both HPC storage systems and machine learning. The enhanced performance on the HPC storage stack will in turn benefit scientific discovery and thus our society. The investigators will integrate research, education, and outreach efforts during the course of this project, including recruiting and retaining of underrepresented students, mentoring graduate and undergraduate students, integrating research findings into curriculum, and publishing and disseminating results.

The planned URL for the project repository is: https://discl.cs.ttu.edu/tuningstorage with links to GitLab repository https://discl.cs.ttu.edu/gitlab. Results and data will be hosted on this site and will be made available by the time of publication. The data will be annotated as appropriate to facilitate interpretation. The principal investigators will strive to maintain the repository as long as possible.

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.