minMLST: machine learning for optimization of bacterial strain typing.

Area of Science:

• Microbiology
• Bioinformatics
• Machine Learning

Background:

• High-resolution microbial strain typing is crucial for clinical applications like outbreak investigation and epidemiological surveillance.
• Core genome multilocus sequence typing (cgMLST) offers high typeability and discriminatory power but faces challenges in implementation, including backward compatibility, nomenclature, typeability, and computational demands.
• Optimizing cgMLST schemes by reducing loci number is beneficial for practical applications.

Purpose of the Study:

• To present minMLST, a novel machine learning methodology for minimizing gene numbers in cgMLST schemes.
• To identify informative gene subsets and analyze the trade-off between gene reduction and typing performance.
• To improve the implementation of cgMLST for better interlaboratory agreement and communication.

Main Methods:

• Developed a machine learning-based methodology named minMLST.
• Implemented gene subset identification and analysis of typing performance versus gene reduction.
• Validated the methodology across eight bacterial species.

Main Results:

• minMLST successfully reduced the number of genes in cgMLST schemes by up to a factor of 10.
• Typing performance remained high, with Adjusted Rand Index values between 0.4 and 0.93 (P < 10-3) across species.
• The method demonstrates significant utility in optimizing cgMLST for bacterial strain typing.

Conclusions:

• minMLST offers an effective approach to optimize cgMLST schemes by reducing loci number while preserving high typing accuracy.
• This optimization is expected to enhance the practical implementation of cgMLST in various settings.
• The minMLST Python package is available for Linux and Windows.