Identifying gut microbiome signatures of type 1 diabetes using machine learning and evolutionary feature selection

Area of Science:

• Microbiome research
• Computational biology
• Immunology

Background:

• Type 1 Diabetes Mellitus (T1D) is increasingly linked to gut microbiome alterations.
• Understanding the impact of taxonomic resolution, feature selection, and machine learning on T1D prediction from microbiome data is crucial.

Purpose of the Study:

• To evaluate microbiome-based prediction of T1D using various taxonomic resolutions and machine learning methods.
• To identify robust and predictive microbial signatures for T1D using Binary Particle Swarm Optimization (BPSO).

Main Methods:

• Analysis of 16S rRNA gene sequencing datasets from two cohorts.
• Microbial feature construction at multiple taxonomic levels and as phylogenetic paths.
• Machine learning model training with cross-validation and cross-cohort validation.
• Feature selection using BPSO for compact and predictive microbial signatures.
• Performance evaluation using AUC, Accuracy, F1 score, and MCC.

Main Results:

• Tree-based models (Random Forest, XGBoost) showed strong predictive performance.
• Family-level features offered a balance of performance and simplicity.
• BPSO identified stable predictive microbial signatures, some linked to inflammation.
• Cross-cohort validation revealed performance limitations, indicating generalization challenges.

Conclusions:

• Machine learning with BPSO feature selection effectively identifies T1D-associated microbial signatures.
• Taxonomic resolution, feature stability, and cross-cohort validation are critical for predictive microbiome modeling.
• Integrating evolutionary feature selection with ML and biological validation can enhance signature robustness and interpretability.