Artificial Intelligence Enabled Experiments and Modeling for Determining Extracellular Vesicle Mechanical Properties Towards Biomarker Technologies

Extracellular vesicles naturally occur in biological fluids. These vesicles and their cargo are a promising source of biomarkers for personalized healthcare, including early cancer detection, health monitoring, and infectious disease. However, current methods for isolating and analyzing these particles are derived from cellular analysis methods, such as size-based sorting, biological targeting, or density-based isolation. For vesicles, these methods are often costly, time consuming, and produce inconsistent results. A key limitation is the poor understanding of vesicle mechanical properties, such as stiffness. Due to their small size and low concentration in biofluids, it is difficult to evaluate mechanical responses such as deformation during analysis. As a result, these mechanical properties are difficult to measure. Their small size also makes it difficult to directly observe individual vesicles using conventional optical microscopy. In this project, new approaches will be developed to quantify the mechanical properties of extracellular vesicles. This work will enable faster and more reliable manipulation of these particles. The project will support applications such as biomarker discovery. The methods in this project combine artificial intelligence–enabled models with state-of-the-art electron microscopy and super-resolution fluorescence microscopy. Together, these tools will provide meaningful measurements of mechanical properties. Additionally, this project will train personnel in advanced biomedical techniques. The trained workforce will support broader impact goals to sustain American leadership in biotechnology. This project will establish a physics-informed artificial intelligence assisted computational framework by utilizing high-resolution experimental imaging data to quantify the mechanical properties of extracellular vesicles. A structural representation based on implicit functions will be developed to capture vesicle geometry and deformation. A reduced-order mechanical model will relate observed shape changes to underlying material properties such as elasticity. Machine learning methods will analyze large imaging datasets, generated through this work along with using previously published data to identify deformation patterns thereby enabling inverse estimation of mechanical parameters for heterogenous vesicle populations and biological fluids. The expected outcomes include robust, high resolution imaging data sets, new algorithms, and methods for extracting mechanical information from images, reliable models for vesicle deformation, and a potentially transformative foundation for mechanobiology by linking nanoscale mechanical properties to biological function. 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: 2534703 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Shaurya Prakash | Institution: OHIO STATE UNIVERSITY, THE, COLUMBUS, OH | Award Amount: $500,000 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2534703 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2534703.html