velvet assembly

[vani s kulkarni]

hi

Am using 'Velvet' assembly for my short read sequence from solexa.
I want to know about the parameters like hash length (K) and N50 in velvet in detail and in the manual these are not explained so good. And I saw the paper by Daniel R. zerbino and Ewan Birney but am not satisfied.

so any one can explain these parameters and how these are used in the decision of good assembly selection? or any web reference is also help full for me.

Thanks in advance....

[bioinfosm]

Can you expand on specific questions?

The N50 length is the length-weighted median contig length. (50% of the nucleotides are situated on a contig longer than n50)

[Melissa]

You mean this http://www.ebi.ac.uk/~zerbino/velvet...th_choice.html

[vani s kulkarni]

Can you expand on specific questions?

The N50 length is the length-weighted median contig length. (50% of the nucleotides are situated on a contig longer than n50)

hi

how to select good assembly based on hash length and N50 in velvet denovo assembler?

[vani s kulkarni]

You mean this http://www.ebi.ac.uk/~zerbino/velvet...th_choice.html

hi
thanks for the link

but I want to know how this hash length helps in selecting good assembly?

[Rao]

Hash length is overlap b/w reads (based on this it makes graph).... if it more you get better assembly....if you keep less, there is a chance of mis-assemblies....I think maximum you can keep 31 (it should be odd number)

[Torst]

Hash length is overlap b/w reads ....I think maximum you can keep 31 (it should be odd number)

The latest version of Velvet now supports hash values (k) greater than 31. This could improve assemblies of longer reads.

[Rao]

Thanks for the update...

[Zigster]

this plot might help. i generate one of these every time I do an assembly and pick a combination of parameters that keep me somewhere in the upper right of that plot. The exact choice depends on the purpose of the assembly.

the axes are in base pairs. the colored strata are cvCut cohorts. The points are kmers, increasing left to right (I have labeled one set for clarity)

increasing the kmer index will demand greater extension specificity and reduce the number of spurious contigs. If your sequencing went well and you have long reads (i.e. 1.5x-2x kmer) increasing the kmer should not be too detrimental to your total coverage (in this plot assembly length in base pairs). If your kmer is too high you will see a big dropoff in coverage without gaining N50.

increasing the cvCut will raise the contig depth threshold and also help discourage the formation of false contigs at the risk of missing areas of low coverage. If your cvCut is too low you will see a big hit to your N50 at all kmer lengths.

[jnfass]

@Zigster: that's awesome ... would you be willing to share your script for graphing? (R, I'd assume?)

[Zigster]

It's perl and R/Sweave glued together with a shell script

I'll try to get something usable up in googlecode soon

[vani s kulkarni]

hi
Thanks for your informations ,now I can try with different options...

[Zigster]

as promised with better plots:

http://code.google.com/p/standardized-velvet-assembly-report/

[lcollado]

Looks great Just checked out the repository ^^ to look into the details

As a more basic R question, do you use ggplots2 to make the boxplots on the axis as on the first figure you posted here? I ran the testInstalledPackage function (tools library) and didn't see one like that.

Thanks

[Zigster]

that was done using scatterplot from the "car" package
ggplot is much more sophisticated, although I have yet to fully grasp the syntax