wtdbg2: A C repository from grassa

Getting Started

git clone https://github.com/ruanjue/wtdbg2
cd wtdbg2 && make
# assemble long reads
./wtdbg2 -t 16 -i reads.fa.gz -fo prefix -L 5000
# derive consensus
./wtpoa-cns -t 16 -i prefix.ctg.lay -fo prefix.ctg.lay.fa

Introduction

Wtdbg2 is a de novo sequence assembler for long noisy reads produced by PacBio or Oxford Nanopore Technologies (ONT). It assembles raw reads without error correction and then builds the consensus from intermediate assembly output. Wtdbg2 is able to assemble the human and even the 32Gb Axolotl genome at a speed tens of times faster than CANU and FALCON while producing contigs of comparable base accuracy.

During assembly, wtdbg2 chops reads into 1024bp segments, merges similar segments into a vertex and connects vertices based on the segment adjacency on reads. The resulting graph is called fuzzy Bruijn graph (FBG). It is akin to De Bruijn graph but permits mismatches/gaps and keeps read paths when collapsing k-mers. The use of FBG distinguishes wtdbg2 from the majority of long-read assemblers.

Installation

Wtdbg2 only works on 64-bit Linux. To compile, please type make in the source code directory. You can then copy wtdbg2 and wtpoa-cns to your PATH.

Wtdbg2 also comes with an approxmimate read mapper kbm, a faster but less accurate consesus tool wtdbg-cns and many auxiliary scripts in the scripts directory.

Usage

Wtdbg2 has two key components: an assembler wtdg2 and a consenser wtpoa-cns. Executable wtdbg2 assembles raw reads and generates the contig layout and edge sequences in a file "prefix.ctg.lay". Executable wtpoa-cns takes this file as input and produces the final consensus in FASTA. A typical workflow looks like this:

./wtdbg2 -t 16 -i reads.fa.gz -fo prefix
./wtpoa-cns -t 16 -i prefix.ctg.lay -fo prefix.ctg.lay.fa

where -t specifies the number of CPU cores (-t 0 to use all processors). When the default doesn't work well, you may need to apply more options briefly explained as follows.

Wtdbg2 combines normal k-mers and homopolymer-compressed (HPC) k-mers to find read overlaps. Option -k specifies the length of normal k-mers, while -p specifies the length of HPC k-mers. By default, wtdbg2 samples a fourth of all k-mers by their hashcodes. For data of relatively low coverage, you may increase this sampling rate by reducing -S. This will greatly increase the peak memory as a cost, though. Option -e, which defaults to 3, specifies the minimum read coverage of an edge in the assembly graph. You may adjust this option according to the overall sequencing depth, too. Option -A also helps relatively low coverage data at the cost of performance. For PacBio data, -L5000 often leads to better assemblies emperically, so is recommended. Please run wtdbg2 --help for a complete list of available options or consult README-ori.md for more help.

The following table shows various command lines and their resource usage for the assembly step:

Dataset	GSize	Cov	Asm options	CPU asm	CPU cns	Real tot	RAM
E. coli	4.6Mb	PB x20	-t32 -L5000	39s	10m34s	29s	1.1G
C. elegans	100Mb	PB x80	-t32 -L5000 -e4	1h00m	5h06m	16m16s	9.5G
Human NA12878	3Gb	ONT x36	-t36 -p19 -AS2 -e2 -L5000	822h28m	115h59m	27h42m	182.1G
Human NA19240	3Gb	ONT x35	-t32 -p19 -AS2 -e2	706h30m	114h45m	27h33m	177.5G
C. elegans	100Mb	PB x80	-t64 -L5000	1h46m	5h27m	14m17s	10.1G
Human CHM1	3Gb	PB x60	-t64 -L10000	186h15m	131h52m	7h41m	265.2G
Axolotl	32Gb	PB x32	-t96 -L5000 -AS2	3076h13m	1180h03m	88h01m	1626.7G

The timing was obtained on three local servers with different hardware configurations. There are also run-to-run fluctuations. Exact timing on your machines may differ.

Limitations

Wtdbg2 doesn't work with reads longer than 0x3FFFF (~256kb). Longer reads will be split into multiple parts.
Wtdbg2 only works with up to 0x3FFFFFF (~64 million) reads. If you have more reads, please filter short or low-quality reads first.

Getting Help

Please use the GitHub's Issues page if you have questions. You may also directly contact Jue Ruan at ruanjue@gmail.com.