Skip to content
/ wtdbg2 Public

Redbean: A fuzzy Bruijn graph approach to long noisy reads assembly


Notifications You must be signed in to change notification settings


Repository files navigation

Getting Started

git clone
cd wtdbg2 && make
#quick start with
./ -t 16 -x rs -g 4.6m -o dbg reads.fa.gz

# Step by step commandlines
# assemble long reads
./wtdbg2 -x rs -g 4.6m -i reads.fa.gz -t 16 -fo dbg

# derive consensus
./wtpoa-cns -t 16 -i dbg.ctg.lay.gz -fo dbg.raw.fa

# polish consensus, not necessary if you want to polish the assemblies using other tools
minimap2 -t16 -ax map-pb -r2k dbg.raw.fa reads.fa.gz | samtools sort -@4 >dbg.bam
samtools view -F0x900 dbg.bam | ./wtpoa-cns -t 16 -d dbg.raw.fa -i - -fo dbg.cns.fa

# Addtional polishment using short reads
bwa index dbg.cns.fa
bwa mem -t 16 dbg.cns.fa sr.1.fa sr.2.fa | samtools sort -O SAM | ./wtpoa-cns -t 16 -x sam-sr -d dbg.cns.fa -i - -fo dbg.srp.fa


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.


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.


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

./wtdbg2 -x rs -g 4.6m -t 16 -i reads.fa.gz -fo prefix
./wtpoa-cns -t 16 -i prefix.ctg.lay.gz -fo prefix.ctg.fa

where -g is the estimated genome size and -x specifies the sequencing technology, which could take value "rs" for PacBio RSII, "sq" for PacBio Sequel, "ccs" for PacBio CCS reads and "ont" for Oxford Nanopore. This option sets multiple parameters and should be applied before other parameters. When you are unable to get a good assembly, you may need to tune other parameters 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 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 -x rs -g4.6m -t16 53s 8m54s 42s 1.0G
C. elegans 100Mb PB x80 -x rs -g100m -t32 1h07m 5h06m 13m42s 11.6G
D. melanogaster A4 144m PB x120 -x rs -g144m -t32 2h06m 5h11m 26m17s 19.4G
D. melanogaster ISO1 144m ONT x32 -xont -g144m -t32 5h12m 4h30m 25m59s 17.3G
A. thaliana 125Mb PB x75 -x sq -g125m -t32 11h26m 4h57m 49m35s 25.7G
Human NA12878 3Gb ONT x36 -x ont -g3g -t31 793h11m 97h46m 31h03m 221.8G
Human NA19240 3Gb ONT x35 -x ont -g3g -t31 935h31m 89h17m 35h20m 215.0G
Human HG00733 3Gb PB x93 -x sq -g3g -t47 2114h26m 152h24m 52h22m 338.1G
Human NA24385 3Gb CCS x28 -x ccs -g3g -t31 231h25m 58h48m 10h14m 112.9G
Human CHM1 3Gb PB x60 -x rs -g3g -t96 105h33m 139h24m 5h17m 225.1G
Axolotl 32Gb PB x32 -x rs -g32g -t96 2806h40m 1456h13m 110h16m 1788.1G

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. The assembled contigs can be found at the following FTP:


  • For Nanopore data, wtdbg2 may produce an assembly smaller than the true genome.

  • When inputing multiple files of both fasta and fastq format, please put fastq first, then fasta. Otherwise, program cannot find '>' in fastq, and append all fastq in one read.

Citing wtdbg2

If you use wtdbg2, please cite:

Ruan, J. and Li, H. (2019) Fast and accurate long-read assembly with wtdbg2. Nat Methods doi:10.1038/s41592-019-0669-3

Ruan, J. and Li, H. (2019) Fast and accurate long-read assembly with wtdbg2. bioRxiv. doi:10.1101/530972

Getting Help

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