Canu FAQ常见问题
链接:Canu FAQ
-
Q: - What resources does Canu require for a bacterial genome assembly(细菌基因组组装)? A mammalian(哺乳类) assembly?
- A:
-
Canu is designed to scale resources(自动测量系统硬件资源) to the system it runs on. It will report if the a system does not meet the minimum requirements for a given genome size.
Typically, a bacterial genome can be assembled in 1-10 cpu hours, depending on coverage (~20 min on 16-cores) and 4GB of ram (8GB is recommended). A mammalian genome (such as human) can be assembled in 10-25K cpu hours, depending on coverage (a grid environment is recommended) and at least one machine with 64GB of ram (128GB is recommended).
-
Q: - What parameters should I use for my genome? Sequencing type? (该用什么参数进行组装?)
- A:
-
By default, Canu is designed to be universal(通用) on a large range of PacBio (C2-P6-C4) and Oxford Nanopore (R6-R9) data. You can adjust parameters to increase efficiency for your datatype. For example, for higher coverage PacBio datasets, especially from inbred(同系交配) samples, you can decrease the error rate (
errorRate=0.013
)(覆盖度足够的话可以降低errorrate,1.3%,从而保证更加精准). For recent Nanopore data (R9) 2D data, you can also decrease the default error rate (errorRate=0.013
).With R7 1D sequencing data, multiple rounds(多轮) of error correction are helpful. This should not be necessary for sequences over 85% identity. You can run just the correction from Canu with the options
-correct corOutCoverage=500 corMinCoverage=0 corMhapSensitivity=high
for 5-10 rounds, supplying the asm.correctedReads.fasta.gz output from round
i-1
to roundi
. Assemble with-nanopore-corrected <your data> errorRate=0.1 utgGraphDeviation=50
-
Q: - How do I run Canu on my SLURM/SGE/PBS/LSF/Torque system? (怎么在集群上运行canu)
- A:
- Canu will auto-detect and configure itself to submit on most grids. If your grid requires special options (such as a partition on SLURM or an account code on SGE, specify it with
gridOptions="<your options list>"
which will passed to the sheduler by Canu. If you have a grid system but prefer to run locally, specify useGrid=false (平时一般都是设置为false) -
Q: - My asm.contigs.fasta is empty, why? (得到的contig文件是空的?)
- A:
-
By default, canu will split the final output into three files:
- asm.contigs.fasta
- Everything which could be assembled and is part of the primary assembly, including both unique and repetitive elements. Each contig has several flags included on the fasta def line:
- asm.bubbles.fasta
- alternate paths in the graph which could not be merged into the primary assembly.
- asm.unassembled.fasta
- reads/tigs which could not be incorporated into the primary or bubble assemblies.
It is possible for tigs comprised of multiple reads to end up in asm.unassembled.fasta. The default filtering eliminates(消除了) anything with < 2 reads, shorter than 1000bp, or comprised of mostly a single sequence (>75%). The filtering is controlled by the contigFilter parameter which takes 5 values.
contigFilter minReads minLength singleReadSpan lowCovSpan lowCovDepth
The default filtering is
2 1000 0.75 0.75 2
. If you are assembling amplified data or viral data, it is possible your assembly will be flagged as unassembled. In those cases, you can turn off the filtering with the parameterscontigFilter="2 1000 1.0 1.0 2"
-
Q: - Why is my assembly is missing my favorite short plasmid X?
- A:
-
The first step in Canu is to find high-error overlaps and generate corrected sequences for subsequent assembly. This is currently the fastest step in Canu. By default, only the longest 40X of data (based on the specified genome size) is used for correction. If you have a dataset with uneven coverage or small plasmids, correcting the longest 40X may not give you sufficient coverage of your genome/plasmid. In these cases, you can set
corOutCoverage=1000
Or any large value greater than your total input coverage which will correct and assemble all input data, at the expense of runtime. This option is also recommended for metagenomic datasets where all data is useful for assembly.
-
Q: - Why do I get only 30X of corrected data?
- A:
-
By default, only the longest 40X of data (based on the specified genome size) is used for correction. Typically, some reads are trimmed during correction due to being chimeric or having erroneous sequence, resulting in a loss of 20-25% (30X output). You can force correction to be non-lossy by setting(数据全部使用、无损输出)
corMinCoverage=0
In which case the corrected reads output will be the same length as the input data, keeping any high-error unsupported bases. Canu will trim these in downstream steps before assembly.
-
Q: - What is the minimum coverage required to run Canu? (最小的覆盖度要求)
- A:
-
We have found that on eukaryotic genomes(真核生物基因组) >=20X typically begins to outperform(胜过) current hybrid methods(混合方法). For low coverage datasets (<=30X) we recommend the following parameters
corMinCoverage=0 errorRate=0.035
For high-coverage datasets (typically >=60X) you can decrease the error rate since the higher number of reads should allow sufficient assembly from only the best subset
errorRate=0.013
However, the above is mainly an optimization for speed and will not affect your assembly continuity.
-
Q: - My genome is AT/GC rich, do I need to adjust parameters? (基因组AT或GC含量偏差比较大怎么设置参数?)
- A:
-
On bacterial genomes, typically no(细菌的不需要设置). On repetitive genomes with AT<=25 or 75>=AT (or GC) the sequence biases the Jaccard estimate used by MHAP. In those cases setting
corMaxEvidenceErate=0.15
has been sufficient to correct for the bias in our testing. In general, with high coverage repetitive genomes(高覆盖率重复的基因组) (such as plants) it can be beneficial to set the above parameter as it will eliminate repetitive matches, speed up the assembly, and sometime improve unitigs.