MV-CVIB: a microbiome-based multi-view convolutional variational information bottleneck for predicting metastatic...

Area of Science:

• Microbiome research
• Oncology
• Bioinformatics

Background:

• Gut microbial imbalances are linked to various diseases, including colorectal cancer (CRC).
• Metastatic colorectal cancer (mCRC) presents a significant clinical challenge due to high mortality and metastasis rates.
• Traditional machine learning models struggle with high-dimensional, small-sample gut microbial data.

Purpose of the Study:

• To develop an effective method for predicting mCRC using gut microbiome data.
• To address the limitations of traditional classification strategies and machine learning models in analyzing complex microbial data.

Main Methods:

• Collected and processed 16S rRNA and abundance data from non-metastatic CRC (non-mCRC) and mCRC patients.
• Proposed a novel microbiome-based multi-view convolutional variational information bottleneck (MV-CVIB) model.
• Employed a disease-disease classification strategy instead of traditional health-disease classification.

Main Results:

• The MV-CVIB model achieved high predictive performance for mCRC, with AUC values exceeding 0.9.
• The model demonstrated consistent and satisfactory predictive accuracy across multiple published CRC gut microbiome datasets.
• Comparative analysis showed MV-CVIB outperforming other state-of-the-art models.

Conclusions:

• MV-CVIB effectively predicts mCRC from gut microbiome data.
• The study highlights the potential of microbiome-based AI models in cancer diagnostics.
• Further analysis elucidated microbial differences between mCRC and non-mCRC, linking them to metastatic properties.