Finimizers: Variable-length bounded-frequency minimizers fork-mer sets

Abstract

AbstractThe minimizer of ak-mer is the smallestm-mer inside thek-mer according to some order relation<of them-mers. Minimizers are often used as keys in hash tables in indexing tasks in metagenomics and pangenomics. The main weakness of minimizer-based indexing is the possibility of very frequently occurring minimzers, which can slow query times down significantly. Popular minimizer alignment tools employ various and often wild heuristics as workarounds, typically by ignoring frequent minimizers or blacklisting commonly occurring patterns, to the detriment of other metrics (e.g., alignment recall, space usage, or code complexity).In this paper, we introducefrequency-bounded minimizers, which we callfinimizers, for indexing sets ofk-mers. The idea is to use an order relation<for minimizer comparison that depends on the frequency of the minimizers within the indexedk-mers. With finimizers, the lengthmof them-mers is not fixed, but is allowed to vary depending on the context, so that the length can increase to bring the frequency down below a user-specified thresholdt. Setting a maximum frequency solves the issue of very frequent minimizers and gives us a worstcase guarantee for the query time. We show how to implement a particular finimizer scheme efficiently using the Spectral Burrows-Wheeler Transform (SBWT) (Alanko et al., Proc. SIAM ACDA, 2023) augmented with longest common suffix information. In experiments, we explore in detail the special case in which we sett= 1. This choice simplifies the index structure and makes the scheme completely parameter-free apart from the choice ofk. A prototype implementation of this scheme exhibitsk-mer localization times close to, and often faster than, stateof-the-art minimizer-based schemes. The code is available athttps://github.com/ElenaBiagi/Finito.