Sensitive and scalable metagenomic classification using spaced metamers, reduced alphabets, and syncmers

Abstract

Accurate taxonomic classification of metagenomic sequencing data is essential for identifying the diverse organisms present in environmental and clinical samples. To address this, we previously developed Metabuli, an alignment-free classifier that bridges the gap between nucleotide-level resolution and protein-level sensitivity. Central to this approach is the metamer, a novel joint DNA and amino acid k-mer concept introduced by Metabuli to classify taxa across varying levels of database representation. In this study, we significantly optimized its core metamer search by incorporating advanced techniques. By introducing spaced metamers and reduced amino acid alphabets to boost sensitivity, we improved both precision and recall by 4.7 and 2.8 percentage points in a species exclusion test. Furthermore, integrating syncmers with spaced metamers halved the reference database size and doubled the classification speed; despite a slight decrease in recall, this configuration continued to outperform state-of-the-art alignment-free tools. Contactjbeom@snu.ac.kr and martin.steinegger@snu.ac.kr