GoldRush: Ade novolong read genome assembler with linear time complexity

Abstract

Motivation: Current state-of-the-art long readde novogenome assemblers follow the Overlap Layout Consensus (OLC) paradigm, an O(n2) algorithm in its naïve implementation. While the most time- and memory-intensive step of OLC —the all-vs-all sequencing read alignment process— was improved and reimplemented in modern long read assemblers, these tools still often require excessive computational memory when assembling a typical 50X human genome dataset. Results: Here we present GoldRush, a de novo genome assembly algorithm with linear time complexity in the number of input long sequencing reads. We tested GoldRush on Oxford Nanopore Technologies datasets with different base error profiles describing the genomes of three human cell lines (NA24385, HG01243 and HG02055),Oryza sativa(rice), andSolanum lycopersicum(tomato). GoldRush achieved NGA50 lengths of 18.3-22.2 Mbp for the three human datasets, with two of the three assemblies having the fewest extensive misassemblies, and NGA50 lengths of 0.3 and 2.6 Mbp for the 373 Mbp and 824 Mbp genomes of rice and tomato, respectively. Further, GoldRush assembled all genomes within a day, using at most 54.5 GB of RAM. These results demonstrate that our algorithm and new assembly paradigm can be used to assemble large genomes de novo efficiently in compute memory space, with resulting assembly contiguity comparable to that of state-of-the-art OLC genome assemblers.

Linear time complexity de novo long read genome assembly with GoldRush

“…The GoldRush, Flye, Redbean, and Shasta genome assemblies generated in this study have been deposited in Zenodo at https://doi.org/10. 5281/zenodo.7884681 53 . The GoldRush genome assemblies generated for the parameter sweep experiments in Supplementary Figs.…”

“…GoldRush (v1.0.0) has been deposited in Zenodo at https://doi.org/10. 5281/zenodo.7884291 54 . GoldRush is available at https://github.com/ bcgsc/goldrush and released under the GPL-3 license.…”

ntLink: A Toolkit for De Novo Genome Assembly Scaffolding and Mapping Using Long Reads

“…Following this gap-filling step, the scaffolds are output in FASTA format. Because the gaps are filled with raw long-read sequence, we recommend polishing the output assembly using long-read polishing tools such as ntEdit+Sealer (Li et al, 2022;Wong et al, 2022), Racon (Vaser et al, 2017) For more information about installing these dependencies, see Support Protocol and Basic Protocol 1, steps 1-2. Instructions for creating a conda environment that can be used for installing protocol-specific dependencies, as described below, are available in Support Protocol.…”

“…ntLink is a lightweight, minimizer‐based long‐read scaffolding tool that was previously published as a central step in the correction and scaffolding pipeline LongStitch (Coombe et al., 2021). More recently, ntLink was integrated as a key step in our de novo long‐read assembler GoldRush (Wong et al., 2022). ntLink uses long‐read evidence to further contiguate draft assemblies from any sequencing technology.…”

“…Although ntLink was developed as a scaffolding tool, the initial minimizer‐guided step of mapping long reads to the draft assembly can also benefit other bioinformatics utilities that require approximate mapping information, including the ntEdit+Sealer polishing component of the GoldRush assembler (Current Protocols article: Li et al., 2022; Paulino et al., 2015; Warren et al., 2019; Wong et al., 2022) and Tigmint‐long (Coombe et al., 2021; Jackman et al., 2018), an assembly correction tool also integrated in the LongStitch pipeline. In Alternate Protocols 1 and 2, we highlight the usage of ntLink as a mapping tool, and, as an example of using ntLink mappings in a different downstream application, showcase how these mappings can be utilized in Tigmint‐long.…”