Enhancing non-invasive colorectal cancer screening with stool DNA methylation markers and light gradient-boosting...

Area of Science:

• Oncology
• Molecular Diagnostics
• Bioinformatics

Background:

• Colorectal cancer (CRC) screening is crucial for early detection and improved outcomes.
• Non-invasive screening methods are highly desirable to increase patient compliance.
• Stool-based DNA methylation markers offer a promising avenue for non-invasive CRC detection.

Purpose of the Study:

• To evaluate the diagnostic effectiveness of stool DNA methylation markers (CNRIP1, SFRP2, VIM) for colorectal cancer (CRC) screening.
• To assess the performance of Fecal Occult Blood Testing (FOBT) in CRC detection.
• To integrate these markers with a Light Gradient Boosting Machine (LightGBM) algorithm for enhanced CRC screening accuracy.

Main Methods:

• Analysis of stool DNA methylation status for CNRIP1, SFRP2, and VIM in 100 samples (50 CRC, 50 normal).
• Parallel performance of Fecal Occult Blood Testing (FOBT).
• Development of a LightGBM machine learning model incorporating methylation markers and FOBT results, with diagnostic performance assessed via ROC curve analysis.

Main Results:

• SFRP2 demonstrated the highest diagnostic accuracy (AUC=0.87), followed by CNRIP1 (AUC=0.83) and VIM (AUC=0.80).
• FOBT showed lower predictive value (AUC=0.67) compared to individual methylation markers.
• The LightGBM model achieved a superior AUC of 0.95, significantly outperforming individual markers, though its sensitivity was 0.78.

Conclusions:

• Stool DNA methylation markers CNRIP1, SFRP2, and VIM show high sensitivity for non-invasive CRC screening.
• Integrating these biomarkers with the LightGBM machine learning algorithm substantially enhances diagnostic accuracy.
• This combined approach represents a promising strategy for early colorectal cancer detection.