Standardized Computational Workflows for Modeling and Inference of RNA Strand-exchange Circuits

This project helps scientists better understand and predict engineered biological systems, including cells designed to sense signals or make useful products, so they can be applied more reliably in the emerging U.S. bioeconomy. Such systems are increasingly important in biotechnology, biomanufacturing, and environmental sensing, but their behavior is often difficult to predict and control in practice. Furthermore, scientists often find that computational models do not match what is observed in experiments, making it difficult to understand system behavior, improve designs, and advance new technologies into practical use. To address this fundamental problem, the project will develop a reproducible computational workflow that uses experimental data and artificial intelligence to improve mechanistic models of engineered biological systems. This work will help promote the progress of science by giving researchers a more systematic way to connect measurements to biological mechanisms, improve the accuracy of predictive models, and uncover previously unrecognized interactions. It also benefits society by supporting more reliable biotechnology, expanding access to reusable computational tools, and training graduate and undergraduate students through research, coursework, and outreach activities. This project will develop a machine-readable computational workflow for improving mechanistic models of engineered biological systems using experimental data. Using genetically encoded RNA strand-exchange circuits as a case study, the project will combine experimental measurements from the National Institute of Standards and Technology with computational modeling to refine models when predictions do not match observations. The work has three technical components: construction and refinement of a baseline mechanistic model using experimental measurements and biologically supported interactions; hybrid residual learning that combines a mechanistic ordinary differential equation model with a compact neural network to extract structured model-experiment discrepancies as interpretable time-resolved rate signals; and verification-driven synthesis that transforms those learned residuals into explicit candidate chemical reaction network mechanisms through piecewise-linear approximation, abstraction-refinement, and correct-by-construction software implemented in Dafny. By integrating mechanistic modeling, artificial intelligence-enabled inference, and formally verified synthesis into a reproducible workflow, the project will advance biotechnology and artificial intelligence priorities while providing a systematic and transferable approach for model refinement, biological discovery, and reusable computational infrastructure in engineering biology. This award is funded by a collaboration between NSF, the National Institute of Standards and Technology, and Schmidt Sciences. 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. NSF Award ID: 2622710 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT,4082CYXXDB NSF TRUST FUND | Principal Investigator: Lukas Buecherl | Institution: Utah State University, LOGAN, UT | Award Amount: $249,955 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2622710 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2622710.html