Identification of Biomarkers for Colorectal Cancer using Machine Learning and Bioinformatics Approaches

Worldwide, colorectal cancer (CRC) is one of the leading causes of cancer-related mortality, and the identification of biomarkers can significantly improve the effectiveness of treatment. In this study, we utilized five different microarray gene expression datasets from Gene Expression Omnibus (GEO) and combined them to identify key potential biomarkers for CRC. Ten feature selection (FS) techniques were used to select key genes from the gene expression matrix. Subsequently, ten machine learning (ML) algorithms (AB, DT, ET, GNB, KNN, LR, MLP, RF, SVC, and XGBoost) were applied to selected key genes to classify cancer and normal samples. Upon evaluating model performance, different classifiers demonstrated strong results using different combinations of FS techniques. In addition, the proposed framework identified the three key genes that appeared in five or more FS techniques. Finally, the prediction model was developed using a three-gene panel (CA7, SLC30A10, and FOXQ1) and demonstrated improved accuracy in classifying cancer and normal samples. Upon evaluation, the GNB classifier had an accuracy of 93.42% and an AUC = 0.90. These three key genes also exhibited significant differences in expression between cancer and normal samples, suggesting that they may contribute to improved clinical outcomes for CRC patients.