A Gene-Based Machine Learning Classifier Associated to the Colorectal Adenoma—Carcinoma Sequence

Colorectal cancer (CRC) carcinogenesis is generally the result of the sequential mutation and deletion of various genes; this is known as the normal mucosa–adenoma–carcinoma sequence. The aim of this study was to develop a predictor-classifier during the “adenoma-carcinoma” sequence using microarray gene expression profiles of primary CRC, adenoma, and normal colon epithelial tissues. Four gene expression profiles from the Gene Expression Omnibus database, containing 465 samples (105 normal, 155 adenoma, and 205 CRC), were preprocessed to identify differentially expressed genes (DEGs) between adenoma tissue and primary CRC. The feature selection procedure, using the sequential Boruta algorithm and Stepwise Regression, determined 56 highly important genes. K-Means methods showed that, using the selected 56 DEGs, the three groups were clearly separate. The classification was performed with machine learning algorithms such as Linear Model (LM), Random Forest (RF), k-Nearest Neighbors (k-NN), and Artificial Neural Network (ANN). The best classification method in terms of accuracy (88.06 ± 0.70) and AUC (92.04 ± 0.47) was k-NN. To confirm the relevance of the predictive models, we applied the four models on a validation cohort: the k-NN model remained the best model in terms of performance, with 91.11% accuracy. Among the 56 DEGs, we identified 17 genes with an ascending or descending trend through the normal mucosa–adenoma–carcinoma sequence. Moreover, using the survival information of the TCGA database, we selected six DEGs related to patient prognosis (SCARA5, PKIB, CWH43, TEX11, METTL7A, and VEGFA). The six-gene-based classifier described in the current study could be used as a potential biomarker for the early diagnosis of CRC.