MADRe: Strain-level metagenomic classification through assembly-driven database reduction

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Strain-level metagenomic classification is crucial for microbial ecology.
• Challenges include unknown sample composition and redundant reference databases.
• Existing methods struggle with accuracy and scalability.

Purpose of the Study:

• To present MADRe, a modular and scalable pipeline for long-read strain-level metagenomic classification.
• To introduce Metagenome Assembly-Driven Database Reduction for improved accuracy.
• To leverage statistical strategies for enhanced database reduction and read reassignment.

Main Methods:

• MADRe combines long-read metagenome assembly with contig-to-reference reassignment.
• An expectation-maximization framework is used for reference database reduction.
• Probabilistic read mapping reassignment is performed on the reduced database.

Main Results:

• MADRe was evaluated on simulated, mock, and real metagenomic datasets.
• The pipeline consistently improved precision by reducing false-positive strain detections.
• Individual steps of MADRe can be used separately, with the classification step performing comparably to other tools.

Conclusions:

• MADRe offers a sensitive and precise approach to strain-level metagenomic classification.
• The Metagenome Assembly-Driven Database Reduction strategy effectively addresses database redundancy.
• MADRe is an open-source tool available for broader research applications.