Gene Expression-Based Colorectal Cancer Prediction Using Machine Learning and SHAP Analysis

Area of Science:

• Genomics
• Bioinformatics
• Oncology

Background:

• Colorectal cancer (CRC) poses a significant global health challenge.
• Accurate and early diagnosis is crucial for effective treatment and improved patient outcomes.
• Genetic markers offer a promising avenue for developing precise diagnostic tools.

Purpose of the Study:

• To develop and validate a robust genetic diagnostic model for colorectal cancer (CRC).
• To identify key genes associated with CRC through integrated bioinformatics analyses.
• To leverage machine learning for predicting CRC risk and diagnosis.

Main Methods:

• Differential gene expression analysis using TCGA database.
• Mendelian randomization analysis with eQTL and CRC outcome data.
• Development and validation of a diagnostic model using nine machine learning algorithms, including XGBoost.
• Gene selection based on differential expression, randomization analysis, and machine learning model importance.

Main Results:

• Identified 3716 differentially expressed genes (DEGs) and 121 CRC-associated genes via Mendelian randomization.
• A final ten-gene signature (RIF1, GDPD5, DBNDD1, RCCD1, CLDN5, ASCL2, IFITM3, IFITM1, SMPDL3A, SUCLG2) was established.
• The XGBoost model achieved an AUC of 0.990, with the final model showing AUCs of 0.9875 (training) and 0.9601 (validation).
• IFITM1 and DBNDD1 were identified as the most influential genes.

Conclusions:

• The gene expression profile in CRC reflects enhanced cell proliferation, metabolism, and immune evasion.
• The developed ten-gene genetic diagnostic model demonstrates strong predictive performance for CRC.
• This model has significant potential for early CRC diagnosis, intervention, and third-tier prevention strategies.