Early detection of pancreatic cancers by novel nanobiosensor-based protease biomarkers using hierarchical decision structure.

Abstract

e16273 Background: There is a critical need to develop fast, reliable, and cost-effective methods for the detection of pancreatic cancer (PC) at the earliest stage to maximize the impact of treatment. To-date, early detection of PC is close to impossible due to the location of the pancreas and the absence of characteristic symptoms in early cancer stages. Methods: Our team of clinicians and scientists has established a fast and reliable nanobiosensor technology that comprises iron/iron oxide nanoparticles attached to a protease or arginase activatable FRET pair (tetrakis (4- carboxyphenyl) porphyrin (TCPP) /cyanine 5.5). Arginase and seven proteases (MMP1, 3, and 9, cathepsin B, and E, urokinase plasminogen activator, and neutrophil elastase) were identified using the Gene Expression Omnibus (GEO) web tool based on their different expression pattern in pancreatic cancer patients, pancreatitis and healthy control subjects. Protease/arginase activities were measured in serum after 1h of incubation. Based on this data, a novel engineering approach to improved early stage detection of pancreatic cancer is reported here. This study was funded by American Cancer Society Institutional Research Grant (IRG‐16‐194‐07), awarded to the University of Kansas Medical Center. Results: In our study, 159 patients were enrolled at KU Cancer Center from 2000-2019, 47 with metastatic PC, 36 with localized PC, 26 pancreatitis and 50 healthy controls using KUCC Biospecimen Repository. The problem of early stage detection of pancreatic cancer can be modeled as a multi-class classification problem. Conventional classification approaches provide at most 77% accuracy for the dataset under consideration. A new hierarchical decision structure with specific feature engineering at each step is introduced here to improve the performance of the classifier. The fundamental premise of this information fusion-based framework involves tailoring the statistically most significant features with appropriate weights to execute an efficient binary classification task at each hierarchical step. An overall accuracy of 95% was achieved for the detection of patients with early pancreatic cancer (see table). Conclusions: Because of the dire survival statistics of pancreatic cancer, detection at the earliest possible time by means of a liquid biopsy will offer the greatest benefit. Novel nanobiosensor based protease biomarkers achieved high accuracy in early detection of pancreatic cancers by applying hierarchical decision structure. Our results need validation in a larger cohort. Predicted true class considering the following combination of classification methods: Step1 – kNN*, step2 – kNN*, step3 – RFC* (Accuracy = 94.97%).[Table: see text]