Using Machine Learning Methods to Develop Diagnostic and Prognostic mRNA Signatures for Pancreatic Cancer in Plasma Small Extracellular Vesicles.

Background: Pancreatic ductal adenocarcinoma (PDAC) is frequently diagnosed in advanced stage due to the absence of effective diagnostic biomarkers. Small extracellular vesicles (sEVs) have recently emerged as potential clinical biomarkers in liquid biopsy. Our study aimed to explore sEV mRNA biomarkers for PDAC diagnosis and identify relevant markers that could guide the prognosis of PDAC patients.

Methods: We analyzed mRNA sequencing of plasma sEVs from 100 participants and employed four machine learning techniques to create and assess the diagnostic models. Partial plasma sEV mRNAs were identified by all four feature extraction methods and used to construct diagnostic model. We also evaluated the predictive value of the model for the survival prognosis of PDAC patients.

Results: Combined with carbohydrate antigen 19-9 (CA19-9), the 4 sEV mRNAs diagnostic signature (d-signature) could well differentiate PDAC patients from non-PDAC individuals, healthy control individuals, and benign pancreatic disease patients with an area under the curve (AUC) of 0.902, 0.971, and 0.845 in training cohort and AUC of 0.803, 0.938, and 0.762 in validation cohort. Furthermore, Cox regression analysis indicated that the score constructed based on the sEV mRNA signature was an independent adverse prognostic factor for survival prognosis of PDAC.

Conclusions: Our study demonstrated the potential utility of the sEV mRNA d-signature in the diagnosis of PDAC via machine learning methods. Simultaneously, the score from this diagnostic model exhibited a significant correlation with adverse outcome in PDAC patients. This provided a novel non-invasive sEV mRNA signature for clinical diagnosis and prognostic evaluation of PDAC patients.