Using Machine Learning Methods to Study Colorectal Cancer Tumor Micro-Environment and Its Biomarkers

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Colorectal cancer (CRC) remains a major global health challenge, necessitating improved early detection and personalized treatment strategies.
• Biomarker discovery is crucial for enhancing diagnostic accuracy and tailoring therapeutic interventions in CRC management.

Purpose of the Study:

• To identify novel gene expression biomarkers for predicting colorectal cancer prognosis.
• To explore the potential of machine learning algorithms in biomarker discovery for CRC.

Main Methods:

• Utilized RNA-seq and gene chip data from TCGA and GEO databases.
• Applied SMOTE for class imbalance, feature selection algorithms (MCFS, Borota, mRMR, LightGBM), and machine learning models (SVM, XGBoost, RF, kNN).
• Employed interpretable machine learning (IML) for network generation and survival analysis.

Main Results:

• Identified five genes (INHBA, FNBP1, PDE9A, HIST1H2BG, CADM3) significantly correlated with CRC patient prognosis.
• Investigated immune cell infiltration and gene mutation rates for the identified biomarkers.
• Generated co-predictive networks to reveal underlying gene relationships.

Conclusions:

• The identified biomarkers hold significant potential for improving early detection and personalized treatment of colorectal cancer.
• These findings may contribute to better clinical outcomes for CRC patients through targeted therapies.