Accelerating Sequence Alignment to Graphs

Abstract

Aligning DNA sequences to an annotated reference is a key step for genotyping in biology. Recent scientific studies have demonstrated improved inference by aligning reads to a variation graph, i.e., a reference sequence augmented with known genetic variations. Given a variation graph in the form of a directed acyclic string graph, the sequence to graph alignment problem seeks to find the best matching path in the graph for an input query sequence. Solving this problem exactly using a sequential dynamic programming algorithm takes quadratic time in terms of the graph size and query length, making it difficult to scale to high throughput DNA sequencing data. In this work, we propose the first parallel algorithm for computing sequence to graph alignments that leverages multiple cores and single-instruction multiple-data (SIMD) operations. We take advantage of the available inter-task parallelism, and provide a novel blocked approach to compute the score matrix while ensuring high memory locality. Using a 48-core Intel Xeon Skylake processor, the proposed algorithm achieves peak performance of 317 billion cell updates per second (GCUPS), and demonstrates near linear weak and strong scaling on up to 48 cores. It delivers significant performance gains compared to existing algorithms, and results in run-time reduction from multiple days to three hours for the problem of optimally aligning high coverage long (PacBio/ONT) or short (Illumina) DNA reads to an MHC human variation graph containing 10 million vertices.Availability-The implementation of our algorithm is available at https://github.com/ParBLiSS/PaSGAL. Data sets used for evaluation are accessible using https://alurulab.cc.gatech.edu/PaSGAL.

Vargas: heuristic-free alignment for assessing linear and graph read aligners

“…We measured performance using GCUPS (giga cell updates per second), which is the number of cells in the dynamic programming matrix computed per second. This standard metric used extensively in the literature normalizes performance to enable comparisons between experiments conducted using reads and references of different lengths (Rahn et al, 2018;Daily, 2016;Jain et al, 2019;Liu and Schmidt, 2014;Liu et al, 2013).…”

“…For a particular row and column position in the matrix, SIMD instructions are used to compute the value for 8-64 queries simultaneously, depending on the vector capacity. The query-parallel strategy has been used for alignment to linear references in SeqAn (Rahn et al, 2018) and to directed acyclic graphs in PaSGAL (Jain et al, 2019).…”

“…When aligning 150bp reads to chromosome 10, SeqAn (Rahn et al, 2018) reported 420 GCUPS using 40 threads on SKX with AVX512BW; on the same dataset, Parasail (Daily, 2016) recorded 74 GCUPS, but used AVX2 instructions that offer only half the throughput per instruction. PaSGAL (Jain et al, 2019), reported 317 GCUPS with 48 threads on SKX with AVX512 instructions when aligning 100bp reads to a graph genome of the 1Mbp Leukocyte Receptor Complex locus including all variants from 1000 Genomes Project Phase 3, not including traceback.…”

“…Rognes (2011) proposed a query-parallel strategy that fills the same element (same row and column) in several dynamic programming matrices at once. This has been shown to be efficient and practical in prior studies (Rahn et al, 2018;Jain et al, 2019). We adopt this approach in Vargas and provide further evidence that it is efficient on modern general-purpose architectures for both linear and graph alignment.…”

“…The most recent human reference genome assembly, GRCh38, includes alternate assemblies for hypervariable loci (Church et al, 2015). Alignment algorithms that account for genetic variants have likewise been proposed (Schneeberger et al, 2009;Vijaya Satya et al, 2012;Huang et al, 2013;Garrison et al, 2018;Rautiainen et al, 2019;Jain et al, 2019).…”