RNAseq analysis
Bioinformatics Analysis Team
McGill University and Genome Quebec Innovation Center
bioinformatics.service@mail.mcgill.ca
•Module #: Title of Module
•2
Why sequence RNA?
• Functional studies
– Genome may be constant but experimental conditions
have pronounced effects on gene expression
• Some molecular features can only be observed at
the RNA level
– Alternative isoforms, fusion transcripts, RNA editing
• Interpreting mutations that do not have an
obvious effect on protein sequence
– ‘Regulatory’ mutations
• Prioritizing protein coding somatic mutations
(often heterozygous)
Modified from Bionformatics.ca
RNA-seq
Isolate RNAs
Generate cDNA, fragment,
size select, add linkers
Samples of interest
Sequence ends
Condition 1
Condition 2
(normal colon) (colon tumor)
100s of millions of paired reads
10s of billions bases of sequence
RNA-seq – Applications
• Gene expression and differential
expression
• Transcript discovery
• SNV, RNA-editing events, variant
validation
• Allele specific expression
• Gene fusion events detection
• Genome annotation and assembly
• etc ...
RNAseq Challenges
• RNAs consist of small exons that may be
separated by large introns
– Mapping splice-reads to the genome is challenging
– Ribosomal and mitochondrial genes are misleading
• RNAs come in a wide range of sizes
– Small RNAs must be captured separately
• RNA is fragile and easily degraded
– Low quality material can bias the data
Modified from Bionformatics.ca
RNA-Seq: Overview
RNA-Seq: Input Data
Input Data: FASTQ
End 1
End 2
Control1_R1.fastq.gz
Control2_R1.fastq.gz
KnockDown1_R1.fastq.gz
Control1_R2.fastq.gz
Control2_R2.fastq.gz
KnockDown1_R2.fastq.gz
KnockDown2_R1.fastq.gz
KnockDown2_R2.fastq.gz
~ 10Gb each sample
@ERR127302.1 HWI-EAS350_0441:1:1:1055:4898#0/1
GGCTCATCTTGAACTGGGTGGCGACCGTCCCTGGCCCCTTCTTGACACCCAGCG
+
4=B@D99BDDDDDDD:DD?B<<=?>6B###########################
@ERR127302.2 HWI-EAS350_0441:1:1:1056:1163#0/1
GAATGAGAGGCCCTCCCCGTGGAGGCATGGTATCCGGCCGAGGGGGCTTAGTCA
Q = -10 log_10 (p)
Where Q is the quality and p is the
probability of the base being incorrect.
QC of raw sequences
QC of raw sequences
low qualtity bases can bias subsequent anlaysis
(i.e, SNP and SV calling, …)
QC of raw sequences
Positional Base-Content
QC of raw sequences
QC of raw sequences
Species composition (via BLAST)
RNA-Seq: Trimming and
Filtering
Read Filtering
• Clip Illumina adapters:
• Trim
trailing
quality
30 between the adapters and reads. First,
Trimmomatic
uses a two-step
approach
to find<
matches
Trimmomatic
a two-step
to 16
findbp)matches
between
thepossible
adaptersposition
and reads.
First,
short sections uses
of each
adapter approach
(maximum
are tested
in each
within
the
short
adaptersh k(maximum
bp) are i tested
in each
possible position
within the
reads.sections
h If thi s sof
or teach
n alignment,
own as te16 ‘seed’
a perfect
or sufficiently
close match,
reads.
h If thi sby
sorthe
t n alignment,
sh kownparameter
as te ‘seed’
i a perfect
or sufficiently
match,
determined
seedMismatch
(see below),
the entire
alignmentclose
between
the
determined
by theisseedMismatch
parameter
(see below),
entire alignment
between
the
read and adapter
scored. This two-step
strategy
results the
in considerable
efficiency
gains,
read
is scored.can
This
results while
in considerable
efficiencyscore
gains,
sinceand
the adapter
seed alignment
be two-step
calculatedstrategy
very quickly,
the full alignment
is
since
the seed
alignment
can be calculated very quickly, while the full alignment score is
calculated
relatively
rarely.
calculated relatively rarely.
The full alignment score is calculated as follows. Each matching base increases the alignment
The
is calculated
as follows.
Each matching
increases
the alignment
scorefull
byalignment
0.6, whilescore
each mismatch
reduces
the alignment
score bybase
Q/10.
By considering
the
•
Filter
for
read
length
≥
32
bp
score
each mismatch
reduces
Q/10.
By considering
the
qualitybyof0.6,
thewhile
base calls,
mismatches
causedthebyalignment
read errorsscore
havebyless
impact.
A perfect match
Trimmomatic
uses
a
two-step
approach
to
find
matches
between
the
adapters
quality
thesequence
base calls,
mismatches
read25errors
less impact.
A perfect
of a 12 of
base
will
score just caused
over 7, by
while
baseshave
are needed
to score
15. As match
such and read
sections
of
each
16value
bp)
aresimple
tested
inusadellab.org
each
possible
of
12 baseshort
sequence
will
score
just7adapter
while
25 bases
are needed
to score
15. As
such . position w
wearecommend
values
of between
-over
15 as7,(maximum
the
threshold
for
alignment
mode.
RNA-Seq: Mapping
Assembly vs. Mapping
mapping
Reference
all vs reference
reads
contig1
assembly
all vs all
contig2
RNA-seq: Assembly vs
Mapping
Reference-based RNA-seq
Ref. Genome or Transcriptome
RNA-seq
reads
De novo RNA-seq
contig1
contig2
Read Mapping
• Mapping problem is challenging:
–
–
–
–
Need to map millions of short reads to a genome
Genome = text with billons of letters
Many mapping locations possible
NOT exact matching: sequencing errors and biological
variants (substitutions, insertions, deletions, splicing)
• Clever use of the Burrows-Wheeler Transform
http://genomebiology.com/2009/10/3/R25
Genome Biology 2009, Volume 10, Issue 3, Article R25
Langmead et al. R25.
increases speed and reduces memory footprint
http://genomebiology.com/2009/10/3/R25
Genome Biology 2009, Volume 10, Issue 3, Article R25
•TableOther
mappers: BWA, Bowtie,
STAR, GEM, etc. Langmead et al. R25.2
1
Bowtie alignment performance versus SOAP and Maq
Table 1
Platform CPU time
Wall clock time Reads mapped per
Bowtie alignment performance versus SOAP and Maq hour (millions)
Bowtie -v 2 Platform
Server CPU
15 mtime
7s
15 mclock
41 s time Reads
33.8 mapped per
Wall
SOAP
91 h 57 m 35 s 91 h 47 m 46 s hour
0.10 (millions)
Peak virtual memory
footprint (megabytes)
Bowtie speed-up Reads aligned (%)
1,149virtual memory
Peak
13,619
footprint (megabytes)
67.4 aligned (%)
Bowtie
speed-up Reads
351×
67.3
Bowtie-v 2 Server
PC
Bowtie
Maq
SOAP
57s s
1516mm741
s s
1517mm41
9117hh5746mm3535s s 9117hh4753mm467 s
29.5
33.8
0.49
0.10
1,353
1,149
804
13,619
-59.8×
351×
71.9
67.4
74.7
67.3
Bowtie
Bowtie
1617mm4158s s
28.8
29.5
1,353
1,353
--
71.9
71.9
Server
PC
1718mm5726s s
TopHat: Spliced Reads
• Bowtie-based
• TopHat: finds/maps to
possible splicing
junctions.
• Important to
assemble transcripts
later (cufflinks)
SAM/BAM
Control1.bam
Control2.bam
SRR013667.1 99 19 8882171 60
76M = 8882214 119
NCCAGCAGCCATAACTGGAAT
GGGAAATAAACACTATGTTCAA
AG
KnockDown1.bam
KnockDown2.bam
~ 10Gb each bam
SRR013667.1 99 19 8882171 60 76M =
8882214 119
NCCAGCAGCCATAACTGGAATGGG
AAATAAACACTATGTTCAAAG
• Used to store alignments
• SAM = text, BAM = binary
Read name
Bases
Base Qualities
Flag
Reference Position
CIGAR
Mate Position
SRR013667.1 99 19 8882171 60 76M = 8882214 119
NCCAGCAGCCATAACTGGAATGGGAAATAAACACTATGTTCAAAGCAGA
#>A@BABAAAAADDEGCEFDHDEDBCFDBCDBCBDCEACB>AC@CDB@>
…
The BAM/SAM format
samtools.sourceforge.net
picard.sourceforge.net
Sort,View, Index, Statistics, Etc.
$ samtools flagstat C1.bam
110247820 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 duplicates
110247820 + 0 mapped (100.00%:nan%)
110247820 + 0 paired in sequencing
55137592 + 0 read1
55110228 + 0 read2
93772158 + 0 properly paired (85.06%:nan%)
106460688 + 0 with itself and mate mapped
3787132 + 0 singletons (3.44%:nan%)
1962254 + 0 with mate mapped to a different chr
738766 + 0 with mate mapped to a different chr (mapQ>=5)
$
RNA-Seq: Alignment QC
RNA-seQc summary
statistics
broadinstitute.org/cancer/cga/rna-seqc
RNA-seQc covergae
graph
Home-made Rscript:
saturation
RPKM Saturation Analysis
RNA-Seq:Wiggle
UCSC: bigWig Track
Format
• The bigWig format is for display of dense,
continuous data that will be displayed in the
Genome Browser as a graph.
• Count the number of read (coverage at each
genomic position:
Modified from http://biowhat.ucsd.edu/homer/chipseq/ucsc.html
RNA-Seq:Gene-level
counts
HTseq:Gene-level
counts
Contro1: 11 reads
Control2: 16 reads
KnockDown1: 4 reads
KnockDown2: 5 reads
TSPAN16
• Reads (BAM file) are
counted for each gene
model (gtf file) using
HTSeq-count:
www-huber.embl.de/users/anders/HTSeq
TSPAN6
TNMD
DPM1
SCYL3
C1orf112
FGR
CFH
FUCA2
…
NFYA
Control1
11
1
435
203
216
2365
6
380
…
888
Control2 KnockDown1
16
4
0
0
743
836
218
416
643
714
5011
2828
1
4
865
431
…
…
827
1674
KnockDown2
5
0
739
352
704
2294
0
523
…
1580
RNA-seq: EDA
gqSeqUtils R package:
Exploratory Data
Analysis
50
MEF_WT_P5 WT
25
MEF_WT_P7 WT
●
MEF_WT_P6 WT
MEF_KO_P6 KO
●
Mutation
V2
MEF_KO_P5 KO
● KO
0
WT
●
MEF_KO_P7 KO
−25
●
MEF_KO_P3 KO
−50
MEF_WT_P3 WT
−50
−25
0
V1
25
RNA-Seq:Gene-level
DGE
Home-made Rscript:
Gene-level DGE
• edgeR and DESeq : Test the effect of
exp. variables on gene-level read
counts
• GLM with negative binomial
distribution to account for biological
variability (not Poisson!!)
DEseq
edgeR
Differential Gene
Expression
SYMBOL
HNRNPC
FAIM2
AC019178
SSC5D
GGT5
EXOC3L4
FOXS1
AQP5
SLC27A3
TIMP4
logFC
-5.26
-4.82
-6.57
-2.95
-3.03
-3.07
-4.02
-3.73
-2.39
-3.29
PValue
9.19E-55
8.02E-29
2.14E-28
2.39E-27
1.03E-26
9.19E-21
1.69E-19
2.82E-19
6.97E-18
1.21E-17
FDR
5.71E-50
2.49E-24
4.42E-24
3.71E-23
1.28E-22
9.51E-17
1.49E-15
2.18E-15
4.81E-14
7.52E-14
counts.C1
12611
191
100
2274
838
359
113
106
736
126
counts.C2 counts.KD1 counts.KD2
12404
244
443
194
11
3
104
1
1
2123
318
276
803
93
117
344
53
34
92
5
8
113
9
8
637
144
129
120
14
12
Downstream Analyses
Pathways/Gene Set (e.g. GOSeq)
Regulatory Networks
Machine Learning / Classifiers
RNA-Seq:Transcriptlevel DGE
Cufflinks: transcipt
assembly
• Assembly: Reports the most
parsimonious set of transcripts
(transfrags) that explain splicing
junctions found by TopHat
Cufflinks: transcript
abundance
• Quantification: Cufflinks implements
a linear statistical model to estimate an
assignment of abundance to each
transcript that explains the observed
reads with maximum likelihood.
Cufflinks: abundance
output
• Cufflinks reports abundances as Fragments
Per Kilobase of exon model per Million
mapped fragments (FPKM)
C: Number of read pairs (fragments) from transcript
N: Total number of mapped read pairs in library
L: number of exonic bases for transcript
• Normalizes for transcript length and lib. size
Cuffdiff: differential
transcript expression
• Cudiff
– Tests for differential expression of a
cufflinks assembly
RNA-Seq:Generate report
Home-made Rscript
Generate report
– Noozle-based html report which describe
the entire analysis and provide a general
set of summary statistics as well as the
entire set of results
Files generated:
– index.html, links to detailed statistics and
plots
For examples of report generated while
using our pipeline please visit our website