Galt's lecture on velvet

• A lot of parameters with velvet to work with
• Easy to install compared to other tools like allpaths
• /campus/BME235/programs/velvet/ is the home directory
  • To install, simply type 'make'
  • Copy binaries
  • Can change the max-kmer maximum size at compilation
    • 31mers are max size for a 64-bit word
  • Can add solid support by adding in a “_de” tag for double-encoded.
  • Velvet is one of the only de-novo assembly tools supporting colorspace reads
• ~velvet/data
  • Contains a simulated genome with simulated reads which can be used for testing the data
  • No documentation describes the files within this directory
• ~velvet/contrib – Contributors directory
  • velvet-estimate-coverage.pl – Estimates expected coverage
    • Maybe reading the stats.txt file which velvet produces
  • VelvetOptimiser-2.1.0
    • Written to try to find the best parameters for assembling the genome
    • Did not work, got stuck on local maximum for exp_cov and cov_cutoff
• ~assemblies/Pog/velvet-assembly1
  • Run velvet without any arguments under the assumption that velvet with estimate some correct parameters
  • README
  • Ran without using any quality values. Velvet does not use quality values even if using fastq input files.
  • Runs in 5-8 minutes
  • First estimated coverage and cutoff values were terrible.
  • Largest contig 713 bp long with an n50 of 27.
  • Assembling a single lane did better with only 1 lane, rather than two gave much better results
• ~assemblies/test/velvet-assembly2
  • Running velvet on simulated 100kb genome with 35bp and 100bp reads.
  • Lowering the kmer value to 21 yielded a genome of 99975/100000 bases
  • Read length and error rate combined with longer kmer lengths may cause problems.
    • A single error in a read may prevent overlapping of K-mers.
    • A kmer is too long then you may not get overlapping
    • A kmer is too short, then you get too much overlapping where it shouldn't occur
  • Starting running on hgw-dev in order to get help from Daniel Zerbino
    • Used less options: -short
    • estimated coverage cutoff 13 which was right
    • saw some larger contigs: 21211max and 5554 n50
    • ran both short and long reads from the same 454 data.
      • Documentation states that short and long reads are the same except for sme tracking data which is potentially expensive
      • 82889max, 28393 n50
    • Combining both lanes of the 454 yielded 142283 max and 50864 n50.
• ~assemblies/Pog/velvet-assembly?
  • Using flowspace data collapsed some nodes
  • -long_multi_cutoff 4 (over default 2)
    • Have to have seen it multiple times before you
    • Showed improvements around 8 multi_cutoff
    • Lowering cov_cutoff
      • max 680241, 224364 n50
      • This is the best so far
  • large kmer values don't show too much of an improvement
• velvet-assembly2
  • utilizes -shortPaired reads
  • Can also utilize unpaired reads as -short
  • may have some problems with reversing f3 primer
  • insert length 2200 (estimated by mapping paired reads onto 454 contigs from newbler)
  • adjusting other paramets did not improve assembly
  • Combining short and long data did not seem to help

Evaluating assembled genomes

• contig-lengths.rdb within assembly directories
• • bottom of bin \t number of bin \t
• take contigs and use various search tools, megablast, blastn, blat-strict-match, differences, differences.short
  • megablast fast but numb (not sensitive)
  • blastn sensitive
  • blat very sensitive. produces psl output.
    • psl longs can get very long
    • labeled table at the top
    • Filter program blat-strict-match selects the best matches.
  • Differences
    • shows differences between assemblies on the pieces that can map