MicroAIbiome: Decoding Cancer Types from Microbial Profiles Using Explainable Machine Learning.

Area of Science:

• Microbiome research
• Computational biology
• Cancer diagnostics

Background:

• Human tissue microbial communities show promise as cancer biomarkers.
• Microbiome data presents challenges for multiclass cancer classification due to compositionality, high dimensionality, and lack of interpretability.

Purpose of the Study:

• To develop an AI pipeline (MicroAIbiome) for classifying five cancer types using genus-level microbial relative abundances.
• To address challenges in microbiome data analysis for cancer classification.
• To enhance model interpretability for uncovering microbial signatures associated with specific cancers.

Main Methods:

• Developed the MicroAIbiome AI pipeline incorporating zero-replacement, centered log-ratio (CLR) transformation, correlation filtering, and recursive feature elimination (RFE).
• Evaluated five machine learning classifiers, with XGBoost showing the best performance.
• Utilized SHapley Additive exPlanations (SHAP) for feature attribution and microbial signature identification.

Main Results:

• MicroAIbiome pipeline successfully classified five cancer types: esophageal carcinoma (ESCA), head and neck squamous cell carcinoma (HNSC), stomach adenocarcinoma (STAD), colon adenocarcinoma (COAD), and rectum adenocarcinoma (READ).
• XGBoost classifier achieved a highest accuracy of 78.23%, surpassing previous studies.
• Identified class-specific microbial signatures, including Corynebacterium for ESCA and Bacteroides for COAD, through SHAP analysis.

Conclusions:

• Compositional data preprocessing and explainable AI are crucial for advancing microbiome-based cancer diagnostics.
• The MicroAIbiome pipeline offers a robust and interpretable approach for multiclass cancer classification using microbiome data.
• Microbiome signatures hold significant potential for early cancer detection and personalized medicine.