Refining the Feasibility of Machine-Learning-Based Diagnostic Model Utilizing Gut Microbiota Analysis for Colorectal...

Area of Science:

• Microbiome research
• Cancer diagnostics
• Machine learning applications

Background:

• A novel colorectal cancer (CRC) diagnostic model was previously developed using machine learning and gut microbiota analysis.
• This study assesses the reproducibility and comparative diagnostic accuracy of this gut microbiota model against the fecal immunochemical test (FIT).
• The practical application potential of the gut microbiota model was also investigated.

Purpose of the Study:

• To evaluate the reproducibility of a machine learning-based gut microbiota diagnostic model for colorectal cancer (CRC).
• To compare the diagnostic accuracy of the gut microbiota model with the fecal immunochemical test (FIT).
• To explore the potential synergistic effect of combining the gut microbiota model with FIT for enhanced CRC detection.

Main Methods:

• Fecal samples were collected from CRC patients and healthy individuals (HI) who had undergone FIT.
• Gut microbiota analysis was conducted using a standardized pipeline.
• Diagnosis was performed using the machine-learning-based gut microbiota model (ml-GMM) and FIT, with consistent cut-off values.

Main Results:

• The machine learning-based gut microbiota model (ml-GMM) demonstrated reproducible diagnostic accuracy.
• True positive rates were 53.1% for ml-GMM and 86.4% for FIT in 81 CRC patients; false positive rates were 7.3% for ml-GMM and 2.4% for 245 HI.
• Combining ml-GMM and FIT achieved a 91.4% positivity rate in CRC patients across all stages, with a 9.4% rate in HI, suggesting a synergistic effect.

Conclusions:

• The diagnostic accuracy of the machine learning-based gut microbiota model (ml-GMM) for colorectal cancer is reproducible.
• Combining ml-GMM with FIT shows potential for detecting more colorectal cancer patients than FIT alone.