Machine-learning-based Analysis Identifies miRNA Expression Profile for Diagnosis and Prediction of Colorectal Cancer: A Preliminary Study

Background: The stage of colorectal cancer (CRC) at the day of diagnosis has the greatest influence on survival rate. Thus, for CRC, which is mainly identified as advanced disease, non-invasive, molecular blood or stool tests could boost the diagnosis and lower mortality. Evaluation of miRNA expression levels in serum of patients diagnosed with CRC is a potential tool in early screening. Screening can be supported by machine learning (ML) as a tool for developing a cancer risk predictive model based on genetic data. Materials and Methods: miRNA was isolated from the serum of 8 patients diagnosed with CRC and 10 patients from a control group matched for age and sex. The expression of 179 miRNAs was determined using a serum/plasma panel (Exiqon). Determinations were conducted using real-time PCR technique on an Applied Biosystems QuantStudio3 device in 96-well plates. A predictive model was developed through the Azure Machine Learning platform. Results: A wide panel of 29 up-regulated miRNAs in CRC were identified and divided into two subgroups: 1) miRNAs with significantly higher serum level in cancer patients vs. controls (24 miRNAs) and 2) miRNAs detected only in cancer patients and not in controls (5 miRNAs). Re-analysis of published miRNA profiles of CRC tumours or CRC exosomes revealed that only 2 out of 29 miRNAs were up-regulated in all datasets including ours (miR-34a and miR-25-3p). Conclusion: Our research suggests the potential role of overexpressed miRNAs as diagnostic or prognostic biomarkers among CRC patients. Such clustering of miRNAs may be a potential direction for discovering new diagnostic panels of cancer (including CRC), especially using ML. The low correspondence between deregulation of miRNAs in serum and tumour tissue revealed in our study confirms previously published reports.