MB-SupCon: Microbiome-based Predictive Models via Supervised Contrastive Learning.

Area of Science:

• Microbiome research
• Metabolomics
• Computational biology
• Disease prediction modeling

Background:

• The human microbiome, comprising trillions of microorganisms, significantly influences host physiology via molecular interactions.
• Integrating microbiome and metabolomics data offers enhanced disease prediction capabilities.
• Existing datasets often lack paired microbiome and metabolomics data, limiting predictive model development.

Purpose of the Study:

• To develop a novel integrative modeling framework, Microbiome-based Supervised Contrastive Learning (MB-SupCon), for microbiome and metabolome data integration.
• To enhance the predictive accuracy of models using only microbiome data by leveraging metabolome information.
• To demonstrate the framework's utility in improving disease prediction and data visualization.

Main Methods:

• Development of the Microbiome-based Supervised Contrastive Learning (MB-SupCon) framework.
• Integration of paired 16S microbiome and metabolomics data from 720 type 2 diabetes patients.
• Application of MB-SupCon to generate microbiome embeddings for improved prediction and data visualization.
• Validation on a large inflammatory bowel disease study dataset.

Main Results:

• MB-SupCon achieved high prediction accuracies for insulin resistance (84.62%), sex (78.98%), and race (80.04%) in type 2 diabetes patients.
• The generated microbiome embeddings facilitated separable clustering of covariate groups in lower dimensions, enhancing visualization.
• MB-SupCon outperformed existing prediction methods in both type 2 diabetes and inflammatory bowel disease studies.
• The framework effectively improved prediction accuracy using only microbiome data.

Conclusions:

• MB-SupCon provides a powerful approach to integrate microbiome and metabolome data for enhanced disease prediction.
• The framework's ability to generate informative microbiome embeddings improves model performance and data interpretability.
• MB-SupCon shows broad applicability and potential for advancing multi-omics disease studies and personalized medicine.