Binning unassembled short reads based on k-mer abundance covariance using sparse coding

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Metagenome assembly is computationally intensive and can miss low-abundance genomes.
• Existing sequence-binning techniques often rely on prior metagenome assembly.
• Large-scale metagenomic datasets present computational challenges for assembly.

Purpose of the Study:

• To develop a scalable pre-assembly binning scheme for microbial genome recovery.
• To enable the recovery of low-abundance genomes from complex metagenomes.
• To leverage sparse dictionary learning for read-level binning.

Main Methods:

• A pre-assembly binning scheme operating on unassembled short reads.
• Application of sparse dictionary learning and elastic-net regularization.
• Joint analysis of microbiomes from the LifeLines DEEP population cohort (n = 1,135).

Main Results:

• Recovery of hundreds of metagenome-assembled genomes, including very low-abundance ones.
• Demonstration of read-level binning at scale using sparse coding techniques.
• Observed read enrichment across six orders of magnitude in relative abundance.

Conclusions:

• Sparse coding enables scalable read-level binning for microbial genome recovery.
• Bin-first strategies can complement assembly-first protocols by targeting distinct genome segregation.
• The method effectively recovers genomes with low relative abundance.