De novo assembly II

Guest lecturer: Stefan Prost, stefan.prost@berkley.edu

• MiSeq has 300 bp reads
• Paired ends read from both directions, so you get one read for each end (may or may not overlap depending on molecule and read size)

	=====> end_1
	________________________
	           end_2 <=====

• Problem: not really sufficient for repetitive regions
  • ends can't be very far apart because Illumina can't handle big molecules
  • not enough info for scaffolding

• Idea: get paired reads that are much farther away (for more scaffolding info)
• Basically, the same idea as paired-ends, except the middle section is missing and the ends are oriented the opposite way.

		<===== end_1
	________________________
	           end_2 =====>

• Genomic DNA > Fragment (2-5 kb) > biotinylate ends > Circularize > Fragment (400-600 bp) > Enrich biotinylated fragments > Ligate adaptors > …
  • Cut DNA, attach a biotin tag to both ends of the target molecule
  • Circularize target molecule
    • This step is hard
  • Circularized molecule is fragmented, fragment with the biotin tag (a.k.a the ends of the original target stuck to each other backwards, with the tags in between) is enriched
  • Then end repair, A-tailing, adapters added, amplification, sequencing
• Dependent on inferring insert size (can be tricky)
• Most companies can get you 8 kb inserts, with skill you can get up to 20 kb
• Complicated process, weird stuff can happen in between
• Important difference between paired ends and mate-pairs: ends are oriented the opposite way.

• Uncommon and expensive, but the gold standard
• Bacterial F-plasmid takes< 40 kb insert size
• Fragment query DNA > DNA into BAC > Transform E. coli with BAC > ~300 kb long reads
• http://www.scq.ubc.ca/wp-content/plasmidtext.gif

• Base quality: Phred scores reported by sequencer.
• Fastq files: fasta files, plus encoded phred scores
  • Need to know if your file has phred33 or phred64 encoding
• Quality for each individual base is not the whole story, the context matters to the signal processing too
• Reads decrease in quality further down the read
  • Depending on the assembler, you might need to trim lower quality reads off the end. Others want untrimmed data
• Pacific Bio doesn’t have GC|AT bias

• FastQC (Most popular tool to tell you about the read library)
  • FastQC reported an issue with our data with kmer count (related to adapter content)
    • This needs to be checked out and diagnosed!
• Preqc
  • Estimates how difficult the assembly will be

G = (pn(1-k+1))/(λ_k)
G = Genome size
pn = proportion of correct reads
k = kmer length
λ_k= mode of the k-kmer count histogram
Simpson 2013, arXiv

• To estimate genome size need to know:i) total number of reads; ii) length of reads; and iii) kmer distribution

• High amount of small kmers are usually errors

Simulated contig length in the k-de Brujin graph can estimate the best kmer to use for assembly. Based on contig length N50