KPop: accurate and scalable comparative analysis of microbial genomes by sequence embeddings

Abstract

Here we introduce KPop, a novel versatile method based on full k-mer spectra and dataset-specific transformations, through which thousands of assembled or unassembled microbial genomes can be quickly compared. Unlike MinHash-based methods that produce distances and have lower resolution, KPop is able to accurately map sequences onto a low-dimensional space. Extensive validation on simulated and real-life viral and bacterial datasets shows that KPop can correctly separate sequences at both species and sub-species levels even when the overall genomic diversity is low. KPop also rapidly identifies related sequences and systematically outperforms MinHash-based methods.