minMLST

minMLST identifies minimal subsets of loci from cgMLST schemes using machine-learning to preserve MLST discriminatory power for bacterial strain typing.

Key Features:

• Machine-Learning Integration: Employs XGBoost, distance-based hierarchical clustering, and SHAP (SHapley Additive exPlanations) to select a minimal subset of genes that preserve strain discrimination.
• Gene Importance Quantification: Quantifies importance of each gene within an MLST scheme to evaluate the trade-off between gene count and typing resolution.
• Optimization of cgMLST Schemes: Reduces the number of loci in cgMLST schemes while addressing backward compatibility, typeability, and computational demands to maintain robust typing performance.
• High Performance with Reduced Genes: Maintains high typing performance with up to a 10-fold reduction in genes, achieving an Adjusted Rand Index of 0.4–0.93 across eight bacterial species with P-values < 10^-3.

Scientific Applications:

• Disease Outbreak Investigation: Provides high-resolution strain typing to support identification and delineation of outbreak-related isolates.
• Microbial Transmission Tracking: Enables tracking of transmission events by distinguishing closely related bacterial strains.
• Epidemiological Surveillance: Improves monitoring and control of bacterial infections through enhanced strain differentiation for surveillance datasets.

Methodology:

Applies XGBoost, distance-based hierarchical clustering, and SHAP to analyze cgMLST gene subsets and evaluate the trade-off between gene reduction and typing performance.