A machine learning approach identified a diagnostic model for pancreatic cancer through using circulating microRNA signatures

Late diagnosis of pancreatic cancer (PC) due to the limited effectiveness of modern testing approaches, causes many patients to miss the chance of surgery and consequently leads to a high mortality rate. Pivotal improvements in circulating microRNA expression levels in PC patients make it possible to diagnose and treat patients at earlier stages. A list of circulating miRNAs was identified in this study using bioinformatics methods in association with pancreatic cancer through analyzing four GEO microarray datasets. The value of top miRNAs was then assessed via using a machine learning method. Taking the advantage of a combinatorial approach consisting of Particle Swarm Optimization (PSO) + Artificial Neural Network (ANN) and Neighborhood Component Analysis (NCA) iterations on a collection of top differentially expressed circulating miRNAs in PC patients, facilitated ranking them by significance. MiRNA's functional analysis in the final index was performed by predicting target genes and constructing PPI networks. Remarkably, the final model consist of miR-663a, miR-1469, miR-92a-2-5p, miR-125b-1-3p and miR-532-5p showed great diagnostic results on investigated cases and the validation set (Accuracy: 0.93, Sensitivity: 0.93, and Specificity: 0.92). Kaplan-Meier survival assessments of the top-ranked miRNAs revealed that three miRNAs, hsa-miR-1469, hsa-miR-663a and hsa-miR-532-5p, had meaningful associations with the prognosis of patients with pancreatic cancer. This miRNA index may serve as a non-invasive and potential PC diagnostic model, although experimental testing is needed. (C) 2020 IAP and EPC. Published by Elsevier B.V. All rights reserved.