Transcript Day 2AM_Intro_to_RNAseq

Introduction to RNAseq
NGS - Quick Recap
• Many applications -> research intent determines
technology platform choice
• High volume data BUT error prone
• FASTQ is accepted format standard
• Must assess quality scores before proceeding
• ‘Bad’ data can be rescued
The Central Dogma of Molecular Biology
Reverse
Transcription
3
RNAseq Protocols
• cDNA, not RNA sequencing
• Types of libraries available:
– Total RNA sequencing (not advised)
– polyA+ RNA sequencing
– Small RNA sequencing (specific size range
targeted)
cDNA Synthesis
Genome-scale Applications
• Transcriptome analysis
• Identifying new transcribed regions
• Expression profiling
• Resequencing to find genetic polymorphisms:
– SNPs, micro-indels
– CNVs
– Question: Why even bother with exome sequencing
then?
What about microarrays??!!!
• Assumes we know all
transcribed regions and
that spliceforms are not
important
• Cannot find anything
novel
• BUT may be the best
choice depending on
QUESTION
Arrays vs RNAseq (1)
• Correlation of fold change between arrays
and RNAseq is similar to correlation between
array platforms (0.73)
• Technical replicates almost identical
• Extra analysis: prediction of alternative
splicing, SNPs
• Low- and high-expressed genes do not
match
RNA-Seq promises/pitfalls
• can reveal in a single assay:
– new genes
– splice variants
– quantify genome-wide gene expression
• BUT
– Data is voluminous and complex
– Need scalable, fast and mathematically principled
analysis software and LOTS of computing resources
Experimental considerations
• Comparative conditions must make biological
sense
• Biological replicates are always better than
technical ones
• Aim for at least 3 replicates per condition
• ISOLATE the target mRNA species you are after
Analysis strategies
• De novo assembly of transcripts:
+ re-constructs actual spliced transcripts
+ does not require genome sequence
easier to work post-transcriptional modifications
- requires huge computational resources (RAM)
- low sensitivity: hard to capture low abundance transcripts
• Alignment to the genome => Transcript assembly
+ computationally feasible
+ high sensitivity
+ easier to annotate using genomic annotations
- need to take special care of splice junctions
# 11
Basic analysis flowchart
Illumina
reads
Re-align
with different
number of
mismatches
etc
un-mapped
Remove
artifacts
AAA..., ...N...
Align
to the
genome
Clip adapters
(small RNA)
un-mapped
Pre-filter:
low complexity
synthetic
"Collapse"
identical
reads
mapped
Count
and
discard
mapped
Assemble:
contigs (exons)
+ connectivity
Filter out low
confidence
contigs
(singletons)
Annotate
# 12
Software
• Short-read aligners
•
BWA, Novoalign, Bowtie, TOPHAT (eukaryotes)
• Data preprocessing
•
Fastx toolkit, samtools
• Expression studies
•
Cufflinks package, R packages (DESeq, edgeR, more…)
• Alternative splicing
•
Cufflinks, Augustus
The ‘Tuxedo’ protocol
• TOPHAT + CUFFLINKS
• TopHat aligns reads to genome and discovers
splice sites
• Cufflinks predicts transcripts present in dataset
• Cuffdiff identifies differential expression
Very widely adopted suite
Read alignment with TopHat
• Uses BOWTIE aligner to align reads to genome
• BOWTIE cannot deal with large gaps (introns)
• Tophat segments reads that remain unaligned
• Smaller segments mostly end up aligning
Read alignment with TopHat (2)
Read alignment with TopHat (3)
• When there is a large gap between segments
of same read -> probable INTRON
• Tophat uses this to build an index of probable
splice sites
• Allows accurate measurement of spliceform
expression
Cufflinks package
• http://cufflinks.cbcb.umd.edu/
• Cufflinks:
– Expression values calculation
– Transcripts de novo assembly
• Cuffdiff:
– Differential expression analysis
Cufflinks: Transcript assembly
• Assembles individual transcripts based on
aligned reads
• Infers likely spliceforms of each gene
• Quantifies expression level of each
Cuffmerge
• Merges transfrags into transcripts where
appropriate
• Also performs a reference based assembly of
transcripts using known transcripts
• Produces single annotation file which aids
downstream analysis
Cuffdiff: Differential expression
• Calculates expression level in two or more
samples
• Expression level relates to read abundance
• Because of bias sources, cuffdiff tries to model
the variance in its significance calculation
FPKM (RPKM): Expression Values
 Fragments Reads Per Kilobase of exon model per
Million mapped fragments
 Nat Methods. 2008, Mapping and quantifying
mammalian transcriptomes by RNA-Seq.
Mortazavi A et al.
C
FPKM = 10 ´
NL
9
C= the number of reads mapped onto the gene's exons
N= total number of reads in the experiment
L= the sum of the exons in base pairs.
Cuffdiff (differential expression)
• Pairwise or time series comparison
• Normal distribution of read counts
• Fisher’s test
test_id gene
locus
ENSG00000000003TSPAN6
ENSG00000000005TNMD
ENSG00000000419DPM1
ENSG00000000457SCYL3
sample_1
sample_2
chrX:99883666-99894988 q1
chrX:99839798-99854882 q1
chr20:49551403-49575092 q1
chr1:169631244-169863408 q1
status
q2
q2
q2
q2
value_1 value_2
NOTEST 0
NOTEST 0
NOTEST 15.0775
OK
32.5626
ln(fold_change) test_stat
p_value significant
0
0
0
1
no
0
0
0
1
no
23.8627 0.459116 -1.39556 0.162848 no
16.5208 -0.678541
15.8186 0
yes
Recommendations
• You can use BOWTIE or BOWTIE2 but
• Use CUFFDIFF2
– Better statistical model
– Detection of truly differentially expressed genes
– VERY easy to parse output file (See example on
course page)