Establishment and evaluation of prediction model for multiple disease classification based on gut microbial data

Area of Science:

• Microbiome research
• Machine learning in medicine
• Disease prediction

Background:

• Gut microbial communities interact with the host immune system.
• Microorganism abundance can serve as disease biomarkers.
• Previous studies focused on individual microorganism markers for disease prediction.

Purpose of the Study:

• To develop a machine learning model for multi-classification of six distinct diseases using gut microbiome data.
• To identify optimal microbial features and classification methods for disease prediction.
• To explore the potential of gut microbiome profiles as simultaneous diagnostic markers.

Main Methods:

• Utilized abundance data of microorganisms across five taxonomic levels from 696 samples.
• Implemented four multi-class classifiers and two feature selection methods (forward selection, backward elimination).
• Evaluated model performance based on taxonomic levels and classifier choice.

Main Results:

• Classification performance improved with lower taxonomic levels, with the genus level yielding the highest accuracy.
• The LogitBoost classifier demonstrated superior performance compared to other models.
• Optimal feature subsets at the genus level were identified using backward elimination.

Conclusions:

• Gut microbiome composition, particularly at the genus level, can effectively distinguish between multiple diseases.
• The LogitBoost model with genus-level features shows promise for simultaneous disease diagnosis.
• Identified microbial features could serve as novel biomarkers for early disease detection.