Abstract
MOTIVATION: Locality-Sensitive Hashing (LSH) is a widely used algorithm for estimating similarity between large datasets in bioinformatics, with applications in genome assembly, sequence alignment, and metagenomics. However, traditional single-metric LSH approaches often lead to inefficiencies, particularly when handling biological data where regions may have diverse evolutionary histories or structural properties. This limitation can reduce accuracy in sequence alignment, variant calling, and functional analysis. RESULTS: We propose Multi-Metric Locality-Sensitive Hashing (M2LSH), an extension of LSH that integrates multiple similarity metrics for more accurate analysis of complex biological data. By capturing diverse sequence and structural features, M2LSH improves performance in heterogeneous and evolutionarily diverse regions. Building on this, we introduce Multi-Metric MinHash (M3Hash), enhancing sequence alignment and similarity detection. As a proof of concept, we present BisHash, which applies M2LSH to bisulfite sequencing, a key method in DNA methylation analysis. Although not fully optimized, BisHash demonstrates superior accuracy, particularly in challenging scenarios like cancer studies where traditional approaches often fail. Our results highlight the potential of M2LSH and M3Hash to advance bioinformatics research. AVAILABILITY AND IMPLEMENTATION: The source code for BisHash and the test procedures for benchmarking aligners using simulated data are publicly accessible at https://github.com/hnikaein/bisHash.