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

Area of Science:

• Genomics
• Bioinformatics
• Oncology

Background:

• Colorectal cancer (CRC) develops through a stepwise genetic alteration process known as the normal mucosa-adenoma-carcinoma sequence.
• Understanding gene expression changes during this sequence is crucial for early detection and intervention.

Purpose of the Study:

• To develop a predictive classifier for the adenoma-carcinoma transition in CRC using gene expression profiles.
• To identify key differentially expressed genes (DEGs) associated with CRC progression and patient prognosis.

Main Methods:

• Utilized microarray gene expression data from 465 samples (normal, adenoma, CRC) from the Gene Expression Omnibus database.
• Applied sequential Boruta algorithm and Stepwise Regression for feature selection, identifying 56 significant DEGs.
• Employed machine learning algorithms (LM, RF, k-NN, ANN) for classification, with k-NN showing the highest accuracy (88.06%).
• Validated models on an independent cohort, confirming k-NN's superior performance (91.11% accuracy).

Main Results:

• Successfully classified normal, adenoma, and CRC tissues using the 56 selected DEGs.
• Identified 17 DEGs exhibiting a clear expression trend across the normal mucosa-adenoma-carcinoma sequence.
• Selected six DEGs (SCARA5, PKIB, CWH43, TEX11, METTL7A, VEGFA) significantly correlated with patient survival data from TCGA.

Conclusions:

• The k-Nearest Neighbors (k-NN) classifier based on 56 DEGs effectively distinguishes between normal, adenoma, and CRC tissues.
• A novel six-gene classifier holds promise as a potential biomarker for the early diagnosis of colorectal cancer.
• Further validation of this gene signature could lead to improved clinical diagnostic tools for CRC.