Screening of serum exosome markers for colorectal cancer based on Boruta and multi-cluster feature selection algorithms.

Background: Early and timely diagnosis benefits the prognosis of patients with colorectal cancer (CRC). The purpose of this study was to explore biomarkers with diagnostic ability in colorectal cancer. Objective: Boruta and multi-cluster feature selection (MCFS) algorithms were employed to analyze the expression data of serum exosome-sourced microRNAs (miRNAs) in CRC patients in the Gene Expression Omnibus (GEO) database to find candidate feature miRNAs that could distinguish between CRC and normal samples. To identify the feature miRNAs with the highest diagnostic ability, different support vector machine (SVM) classifiers were constructed, and the SVM classifier with the highest F1 score was selected based on IFS curve. To validate the clinical application value of the classifier, serum samples from 32 CRC patients and 19 healthy individuals were collected as external validation sets, and serum exosomes were extracted for quantitative real-time polymerase chain reaction (qRT-PCR) analysis. The data were imported into the model to verify the performance of the model. Result: After feature selection by Boruta and MCFS algorithms, the first five candidate miRNAs (miR-21, miR-193b, miR-23a, miR-575, and miR-610) with sufficient ability to distinguish sample types were identified. In a series of classifiers constructed, the SVM classifier composed of the first four feature miRNAs (miR-21, miR-193b, miR-23a, and miR-575) was determined to have the best classification effect. qRT-PCR results of serum exosome miRNAs in clinical samples demonstrated that the expression of miR-21, miR-193b, and miR-23a in serum exosomes from CRC patients was significantly higher than that in normal samples, while that of miR-575 was significantly lower than that in normal samples. Subsequently, the receiver operating characteristics (ROC) curve of the diagnostic model based on four feature miRNAs was plotted. According to the results, the area under concentrations curves (AUC) value of the diagnostic model was 0.854, which suggested that the predictive performance of the model built on miR-21, miR-193b, miR-23a, and miR-575 was effective enough to distinguish healthy subjects from CRC patients. In addition, the expression of miR-21, miR-193b, miR-23a, and miR-575 was closely related to the tumor size, stage, and the presence of distant metastasis in CRC patients. Conclusion: MiR-21, miR-193b, miR-23a, and miR-575 may be a new potential biomarker combination for the diagnosis of CRC. Clinical biopsy combined with these miRNA biomarkers is expected to promote the early diagnosis of CRC, thus optimizing the prognosis of CRC patients. [ABSTRACT FROM AUTHOR]