Common Sequence Analysis Work Flows

Martin Morgan, Sonali Arora
February 3, 2015

RNA-seq

See the lecture notes and lab.

RNA-seq differential expression of known genes

• Simplest scenario
• Experimental design: simple, replicated; track covariates and be aware of batch effects
• Sequencing: moderate length and number of reads; single or paired-end (though probably paired-end).
• Alignment: basic splice-aware aligner, e.g., Bowtie2, STAR. Viable Bioconductor approaches: Rsubread, Rbowtie (especially via the QuasR package).
• Reduction: GenomicRanges::summarizeOverlaps() or external tools, using gene model from TxDb.* package or GFF / GTF files. End result: matrix of counts.
• Analysis: DESeq2, edgeR, and additional software.

RNA-seq differential expression of known and novel transcripts

• Popular non-R work flow: Rbowtie2, tophat, cufflinks, cuffdiff.

• Biocondutor options

  • DEXSeq: differential exon use.
  • Rsubread::subjunc() for aligning without requiring known gene models.
  • cummeRbund: working with cufflinks output.

Single-cell expression

• monocle

ChIP-seq

See my recent slides outlining ChIP-seq and relevant Bioconductor software.

• Experimental design / wet lab: important to effectively enrich genomic DNA via ChIP, otherwise hard to distinguish signal peaks from background
• Sequencing: moderate length and number of single-end reads very adequate.
• Alignment: Basic aligners sufficient

• Reduction

  • External software; many tools depending on application, e.g., MACS.
  • Product: BED and / or WIG files of called peaks

• Analysis & Comprehension

  • ChIPQC for quality control.
  • rtracklayer to input BED and WIG files to standard Bioconductor data structures.
  • ChIPpeakAnno, ChIPXpres for annotating peaks in relation to genes.
  • DiffBind to assess differential representation of peaks in a designed experiment.
  • AnnotationHub for accessing (some) consortium-level summary data.

Copy Number

Experimental design

• Duplications or deletions larger than 1 kb
• Germ line (primarily diploid genome, homogeneous sample, integer copy numbers) or somatic variants?
• Tumor / normal pairs?

Assays

• aCGH, SNP, and other arrays (CGHbase, crlmm, CopyNumber450K)
• Low or high-coverage; exome or whole-genome sequencing.

Reduction

• Bin and count. GC and other (e.g., exon length) correction. Easily and efficiently done with, e.g., GenomicRanges::tileGenome() and GenomicFiles.
• Segment – circular binary segmentation (often via DNAcopy), HMM

Analysis & comprehension

• 45 packages tagged with “CopyNumberVariation” in biocViews; also terms “DNASeq”, “ExomeSeq”, “WholeGenome”
• Represent duplicated regions as genomic ranges; integrates very easily in Bioconductor annotation work flows.

Variants

See Michael Lawrence's variant calling with VariantTools. and Val Obenchain's manipulation and annotation of called variants with VariantAnnotation.

• Sequencing: requires high-quality reads with high per-nucleotide depth of coverage – longer, paired-end sequencing.

• Alignment: requires effective aligners; BWA, GMAP, …

  • gmapR wraps the GMAP aligner in R.

• Reduction: typically to VCF files summarizing variants and / or population-level variation. GATK and other non-R tools commonly used.

  • VariantTools includes facilities for calling variants.
  • h5vc targets a different intermediate step: summarize base counts at each position in the genome; use this as a starting point for calling variants, and to evaluate false positives, etc.

• Analysis & comprehension

  • VariantAnnotation, ensemblVEP for querying / inputting VCF files, and for annotation of variants (“is this a coding variant?”, etc.).
  • SomaticSignatures for working with somatic signatures of single-nucleotide variants.

Epigenomics

See the short introduction and lab centered around Illumina 450k methylation arrays and the minfi package.

• Analysis & comprehension: bsseq, BiSeq for processing and analysis; bumphunter as basic tool for identifying CpG features.

Microbiome

• Experimental design: typically population-level surveys with moderate (10's-100's) of samples.
• Wet lab & sequencing: often target phylogenetically-informative genes, requiring longer (overlapping) paired-end reads. Many existing studies used 454 technology, which has a different sequencing error model than Illumina (e.g., homopolymers are a common error, instead of trailing nucleotide quality deterioration).

• Reduction: Pre-processing (e.g., knitting together overlapping paired-end reads) and taxonomic classification / placement in third-party software, e.g., QIIME, pplacer. End result: count table summarizing represenation of distinct taxa in each sample.

  • rRDP provides an R / Bioconductor interface to the RDP classifiere.

• Analysis: R / Bioconductor and many insights from microarray / RNA-seq analysis well suited to count table, but common pipelines have re- or dis-invented the wheel.

  • phyloseq provides very nice tools for general analysis.