Practical combinations of repetition-aware data structures

TL;DR

This paper investigates combining various repetition-aware data structures, like RLBWT, Lempel-Ziv factors, and compact directed acyclic word graphs, to improve indexing of highly-repetitive string collections in practical scenarios.

Contribution

It introduces practical variants that combine multiple measures of repetition, demonstrating their effectiveness and ease of implementation for indexing highly-repetitive data.

Findings

Combined data structures are space-efficient and scalable.

Variants outperform single-measure approaches on real datasets.

Practical implementations show competitive performance.

Abstract

Highly-repetitive collections of strings are increasingly being amassed by genome sequencing and genetic variation experiments, as well as by storing all versions of human-generated files, like webpages and source code. Existing indexes for locating all the exact occurrences of a pattern in a highly-repetitive string take advantage of a single measure of repetition. However, multiple, distinct measures of repetition all grow sublinearly in the length of a highly-repetitive string. In this paper we explore the practical advantages of combining data structures whose size depends on distinct measures of repetition. The main ingredient of our structures is the run-length encoded BWT (RLBWT), which takes space proportional to the number of runs in the Burrows-Wheeler transform of a string. We describe a range of practical variants that combine RLBWT with the set of boundaries of the Lempel-Ziv 77 factors of a string, which take space proportional to the number of factors. Such variants use, respectively, the RLBWT of a string and the RLBWT of its reverse, or just one RLBWT inside a bidirectional index, or just one RLBWT with support for unidirectional extraction. We also study the practical advantages of combining RLBWT with the compact directed acyclic word graph of a string, a data structure that takes space proportional to the number of one-character extensions of maximal repeats. Our approaches are easy to implement, and provide competitive tradeoffs on significant datasets.