NANOFLUIDIC ISOLATION AND QUANTIFICATION OF SPECIFIC EXTRACELLULAR VESICLES AND MACHINE LEARNING ANALYSIS TO AID CLINICAL DECISION-MAKING

Extracellular vesicles (EVs) hold immense promise as reservoirs of critical biomarkers for diagnosing and prognosing various diseases. However, leveraging their potential in clinical settings faces significant obstacles, including precise detection, quantification, and isolation of specific EV subgroups, as well as complex analysis of liquid biopsy (LB) datasets derived from EV cargoes. In response to these challenges, particularly in the realms of oncology and neurology, our research employs a comprehensive strategy.Building on prior work on high throughput nanomagnetic EV isolation, we have significantly improved the specificity when isolating EV subpopulations based on surface markers. We focused on the analysis of EV-miRNAs and, when combined with the evaluation of circulating protein biomarkers, generated predictive insights relevant to a range of clinical questions. Specifically, we isolated neuron-derived EVs to diagnose and prognose neurological outcomes post-cardiac-arrest brain injury, and EVs from non-small-cell lung cancer cells to predict immunotherapy responses. To sensitively quantify EV subpopulations, our Droplet-based Extracellular Vesicle Analysis (DEVA) achieved limits of detection (LOD) and quantification (LOQ) that are more than 100 times lower than those achieved by the gold standard conventional ELISA. When applied to the quantification of EVs expressing surface Programmed Death-Ligand 1 (PD-L1), in both cell culture samples and human plasma, DEVA maintained the low LOD and LOQ and agreed with gold standard method, offering potential in immunotherapy outcome prediction and melanoma prognosis. To streamline LB dataset analysis, we developed a web-based, automated machine learning tool validated across 11 datasets. Its performance rivals or exceeds custom-designed algorithms, with capacity for improvement through historical data assimilation. A differential privacy algorithm which manages information feedback to users, was integrated to counter overfitting. These advancements represent a concerted effort to explore EVs' diagnostic and prognostic potential in clinical scenarios, with a focus on patient outcomes in cancer and brain injury. Our work is dedicated to cultivating more accurate and targeted biomarker analyses, addressing critical challenges in EV research for enhanced clinical utility.

Author dc:creator

• Shen, Hanfei

Advisors dc:contributor.advisor

• Issadore, David, A
• Tsourkas, Andrew

• Engineering

Language dc:language.iso

en

Repository record dc:identifier.uri

https://repository.upenn.edu/handle/20.500.14332/60109

OAI identifier oai:identifier

oai:repository.upenn.edu:20.500.14332/60109