Serum microRNA panel excavated by machine learning as a potential biomarker for the detection of gastric cancer

Early detection of gastric cancer (GC) is crucial to improve the therapeutic effect and prolong the survival of patients. MicroRNAs (miRNAs) are a group of small non-protein-coding RNAs that function as repressors of diverse genes. We aimed to identify a microRNA panel in the serum of patients to predict GC non-invasively with high accuracy and sensitivity. Using six types of classifiers, we selected three markers (miR-21-5p, miR-22-3p and miR-29c-3p) from a published miRNA profiling study (GSE23739) which was treated as a training set. The values of the area under the receiver operating characteristic (ROC) curves (AUCs) were 0.9437, 0.9456 and 0.9563 in the three classifiers [Compound covariate classifier, Diagonal linear discriminant analysis (DLDA) classifier and Support vector machine classifier], respectively. Then the panel was validated further in another two miRNA profiles in GEO (Gene Expression Omnibus) databases (GSE26595, GSE28700) with high AUC values as well. Next, we found that the serum levels of miR-21 were significantly higher in GC patients than levels in healthy controls by quantitative reverse transcription-polymerase chain reaction (qRT-PCR) for confirmation, which was opposite to the serum levels of miR-22 and miR-29c (all P<0.0001). Finally, using bioinformatic tools, their biological mechanisms were elucidated by their predicted targets: Sp1 (miR-21) and PTEN (miR-22 and miR-29c). This miRNA panel is a non-invasive and potential biomarker for GC.