Bioconductor Workflows Following Fast, Lightweight RNA-seq Quantifiers

Bioconductor Workflows Following Fast, Lightweight RNA-seq Quantifiers

Michael Love @mikelove Bioc2016 June 25, 2016 this talk: http://mikelove.github.io/bioc2016

A typical gene-level RNA-seq pipeline

• Align reads (15-30 min, 2-5 Gb file)
• Count reads in genes (15-30 min)
• Stats packages for inference:
  • DESeq2, edgeR, limma-voom, etc.
• Counts of reads: precision of log2FC

Criticisms of the standard count-based pipeline

• Counts scale with feature length Trapnell et al 2013
• Learning bias (e.g. positional) easier if you know the source

Criticisms of the standard count-based pipeline II

• Discards reads that cannot be uniquely assigned to genes Robert & Watson 2015
• In many cases, we can identify the source (or a set of similar sources)

Criticisms of the standard count-based pipeline III

• Generates large intermediate file with exact alignments which you may not need

New, fast transcript quantifiers

• Sailfish, Salmon, kallisto
• Not exact base-by-base alignments
• Rough location of read within a set of txs
• Few min / file, small memory req'd
• Output relative abundance per tx

Diagrams from: Sailfish, kallisto, RapMap pubs

Using with gene DE

http://bioconductor.org/packages/tximport & F1000Research

• Sum transcript-level estimated counts to gene-level
• Collapse isoform uncertainty
• Probabilistically assign (genomic) multimapping reads, increase sensitivity

Using with gene DE

• Calculate an offset that accounts for changes in average transcript length across samples

\textrm{ATL}_{gs} \equiv \sum_{i \in g} \theta_{is} \bar{l}_{is}, \quad \sum_{i \in g} \theta_{is} = 1

sample .............................. s gene ................................... g isoform ............................. i effective length ............ \bar{l} percent abundance ... \theta

Gene-level and tx-level complementary

Packages for DTE: cuffdiff, BitSeq, EBSeq, sleuth (w/ kallisto), ...

Packages for DEU/DTU: DEXSeq, cuffdiff, MISO, diffSplice, rMATS, ...

Ex: Roadmap tissues

Code on GitHub

Run Salmon on 37 FASTQ: ~4 min / file

# 25 seconds to import and summarize
txi <- tximport(files, type="salmon", tx2gene=tx2gene, reader=read_tsv)
# build DESeq2 object
dds <- DESeqDataSetFromTximport(txi, samples, ~tissue)
# 4 seconds to variance stabilize
vsd <- vst(dds)
# exploratory data analysis
plotPCA(vsd, "tissue")
# differential expression
dds <- DESeq(dds)
res <- results(dds)

Ex: Roadmap tissues

What's next

• tximportMeta wrapping tximport
• Should tell us about samples, transcriptome, what software was used, what options

This work in collaboration with

• Charlotte Soneson @CSoneson (UZH)
• Mark Robinson @markrobinsonca (UZH)

• Soneson, C., Love, M.I., Robinson, M.D. Differential analyses for RNA-seq: transcript-level estimates improve gene-level inferences. F1000Research, Dec 2015.

Support from

• Rafael Irizarry @rafalab (DFCI & HSPH)
• NIH Cancer Training Grant