Taxonomy dimension reduction for colorectal cancer prediction

Area of Science:

• Microbiology
• Bioinformatics
• Oncology

Background:

• Colorectal cancer (CRC) is a leading cause of cancer-related deaths globally.
• Early diagnosis and treatment are crucial for improving patient outcomes.
• Alterations in gut microbial communities are associated with CRC development, suggesting their potential as diagnostic biomarkers.

Purpose of the Study:

• To identify key microbial features for predicting colorectal cancer (CRC).
• To evaluate the effectiveness of various feature selection methods for microbial data in CRC prediction.
• To develop a robust model for early CRC detection using gut microbiome data.

Main Methods:

• Operational taxonomic units (OTUs) representing microbial communities were selected.
• Feature selection involved single methods, correlation-based feature selection, and maximum relevance-maximum distance (MRMD 1.0 and MRMD 2.0).
• Classifiers including random forest, naïve Bayes, and decision trees were employed on training and test sets for evaluation.

Main Results:

• The proposed feature selection methods demonstrated superior performance compared to hierarchical feature engineering.
• The combination of correlation-based feature selection and MRMD 2.0 achieved the best prediction accuracy on the CRC2 dataset.
• Specific microbial features were identified as important predictors of colorectal cancer.

Conclusions:

• Gut microbiome analysis, coupled with advanced feature selection techniques, offers a promising avenue for non-invasive colorectal cancer prediction.
• The integrated feature selection approach (correlation-based feature selection + MRMD 2.0) provides a robust and effective strategy for biomarker discovery in CRC.
• The developed methods and dataset are valuable resources for further research in microbiome-based cancer diagnostics.