MetaCache: context-aware classification of metagenomic reads using minhashing

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Metagenomic shotgun sequencing is widely used for microbiome and environmental studies.
• Accurate taxonomic read classification is crucial but computationally challenging.
• Existing tools face limitations in runtime, memory, or accuracy with large datasets.

Purpose of the Study:

• Introduce MetaCache, a novel software for metagenomic read classification.
• Address the computational challenges of high-throughput sequencing data analysis.
• Provide a memory-efficient and accurate alternative to current read classifiers.

Main Methods:

• Utilizes the big data technique minhashing for read classification.
• Performs context-aware classification by analyzing k-mer subsamples.
• Compares reads against locally constrained regions of reference genomes.

Main Results:

• MetaCache demonstrates significantly lower memory consumption than Kraken and CLARK.
• Achieves highly competitive sensitivity and precision at comparable speeds.
• Successfully builds a database from 140 billion bases of reference genomes using only 62 GB RAM, while other tools fail.
• Classification accuracy improves with increased reference data.

Conclusions:

• MetaCache offers a memory-efficient and accurate solution for metagenomic read classification.
• Its performance scales well with larger reference datasets.
• Represents a significant advancement in analyzing large-scale metagenomic data.