A Machine Learning Method for a Blood Diagnostic Model of Pancreatic Cancer Based on microRNA Signatures

This study aimed to construct a blood diagnostic model for pancreatic cancer (PC) using miRNA signatures by a combination of machine learning and biological experimental verification. Gene expression profiles of patients with PC and transcriptome normalization data were obtained from the Gene Expression Omnibus (GEO) database. Using random forest algorithm, lasso regression algorithm, and multivariate cox regression analyses, the classifier of differentially expressed miRNAs was identified based on algorithms and functional properties. Next, the ROC curve analysis was used to evaluate the predictive performance of the diagnostic model. Finally, we analyzed the expression of two specific miRNAs in Capan-1, PANC-1, and MIA PaCa-2 pancreatic cells using qRT-PCR. Integrated microarray analysis revealed that 33 common miRNAs exhibited significant differences in expression profiles between tumor and normal groups (P value < 0.05 and |logFC| > 0.3). Pathway analysis showed that differentially expressed miRNAs were related to P00059 p53 pathway, hsa04062 chemokine signaling pathway, and cancer-related pathways including PC. In ENCORI database, the hsa-miR-4486 and hsa-miR-6075 were identified by random forest algorithm and lasso regression algorithm and introduced as major miRNA markers in PC diagnosis. Further, the receiver operating characteristic curve analysis achieved the area under curve score > 80%, showing good sensitivity and specificity of the two-miRNA signature model in PC diagnosis. Additionally, hsa-miR-4486 and hsa-miR-6075 genes expressions in three pancreatic cells were all up-regulated by qRT-PCR. In summary, these findings suggest that the two miRNAs, hsa-miR-4486 and hsa-miR-6075, could serve as valuable prognostic markers for PC.