What it does

Velvet is a de novo genomic assembler specially designed for short read sequencing technologies, such as Solexa or 454, developed by Daniel Zerbino and Ewan Birney at the European Bioinformatics Institute (EMBL-EBI), near Cambridge, in the United Kingdom.

Velvet currently takes in short read sequences, removes errors then produces high quality unique contigs. It then uses paired-end read and long read information, when available, to retrieve the repeated areas between contigs.

Read the Velvet documentation for details on using the Velvet Assembler.

Velveth

Velveth takes in a number of sequence files, produces a hashtable, then outputs two files in an output directory (creating it if necessary), Sequences and Roadmaps, which are necessary to velvetg.

Hash Length

The hash length, also known as k-mer length, corresponds to the length, in base pairs, of the words being hashed.

The hash length is the length of the k-mers being entered in the hash table. Firstly, you must observe three technical constraints:

# it must be an odd number, to avoid palindromes. If you put in an even number, Velvet will just decrement it and proceed.
# it must be below or equal to MAXKMERHASH length (cf. 2.3.3, by default 31bp), because it is stored on 64 bits
# it must be strictly inferior to read length, otherwise you simply will not observe any overlaps between reads, for obvious reasons.

Now you still have quite a lot of possibilities. As is often the case, it's a trade- off between specificity and sensitivity. Longer kmers bring you more specificity (i.e. less spurious overlaps) but lowers coverage (cf. below). . . so there's a sweet spot to be found with time and experience. We like to think in terms of "k-mer coverage", i.e. how many times has a k-mer been seen among the reads. The relation between k-mer coverage Ck and standard (nucleotide-wise) coverage C is Ck = C # (L - k + 1)/L where k is your hash length, and L you read length. Experience shows that this kmer coverage should be above 10 to start getting decent results. If Ck is above 20, you might be "wasting" coverage. Experience also shows that empirical tests with different values for k are not that costly to run!

Input Files

Velvet works mainly with fasta and fastq formats. For paired-end reads, the assumption is that each read is next to its mate read. In other words, if the reads are indexed from 0, then reads 0 and 1 are paired, 2 and 3, 4 and 5, etc.

Supported file formats are:

fasta
fastq
fasta.gz
fastq.gz
eland
gerald

Read categories are:

short (default)
shortPaired
short2 (same as short, but for a separate insert-size library)
shortPaired2 (see above)
long (for Sanger, 454 or even reference sequences)
longPaired