NESCENT : NGS : Measuring expression Jen Taylor Bioinformatics Team CSIRO Plant Industry.

NESCENT : NGS : Measuring expression

Jen Taylor

Bioinformatics Team

CSIRO Plant Industry

CSIRO. Nescent August 2011 - Measuring Expression

Measuring Expression• What & Why

• What is expression and why do we care?

• How• Platforms / Technology

• Closed approaches – Microarray• Open approaches - Sequencing

• Experimental Design

• Analysis• Biases• Bioinformatics• Statistical Issues and Analysis

• In action• Workshop – Detection of Differential Expression• Case Studies in Plant functional genomics

What is expression / transcriptome ?

rRNAtRNA

siRNAmicroRNA

tasiRNA lncRNA

Commemorative stained glass window for F.C. Crick, designed by Maria McClafferty.(Photograph: Paul Forster)

Gonville & Caius College, Cambridge, UK.

Beyond the Genome:

Human Genome sequencing begins in earnest

“Mapping the Book of Life”

2000 - First Draft

2003 - Essential Completion

= approx 140, 000 genes

= 30, 000 – 40,000 genes ??

= 24, 195 genes !!!???

“The failure of the human genome”

“despite more than 700 genome-scanning publications and nearly $100bn spent, geneticists still had not found more than a fractional genetic basis for human disease “

Manolio et al., Nature, 2009

“The most likely explanation for why genes for common diseases have not been found is that, with few exceptions, they do not exist.

…., if inherited genes are not to blame for our commonest illnesses, can we find out what is? “

Guardian, 2011

Commemorative stained glass window for F.C. Crick, designed by Maria McClafferty.(Photograph: Paul Forster)

Gonville & Caius College, Cambridge, UK.

Beyond the Genome:

Gene Number ≠ Complexity

lexityRegulation

Transcriptome

Why the expression ?

High-throughput friendly

Context dependent

Regulatory

network

Predicts Biology

Transcriptome

Genome

Proteome

**Li et al., 2004

Measuring Expression ?

Parts Description• Function?

• Interconnectedness?

Comparisons• Population - level• Between genomes

What are important members of a transcriptome?

mRNA• polyadenylated, coding• alternatively spliced

Noncoding RNA (small RNA)• varying lengths, functions (18 – 32 bases)• microRNA, siRNA, piRNA, tasiRNA, long non-coding RNA

“Dark” RNA• transcription outside of annotated genes • Non-polyadenylated

Anti-sense transcription

How does the transcriptome vary to give rise to phenotype ?

Changes in Abundance• Abundance = Rate of Transcription – Rate of Decay

Changes in Function• Availability for function – polyadenylation, silencing, localisation• Suitability for function – alternate splicing

How to measure Expression

PLATFORMS / TECHNOLOGY

Measuring Expression : platforms

• Closed systems – microarray• Probes immobilised on a substrate profile target species in the

transcriptome

Single and two colour arrays

Labelling

Two colour

Control

Experimental

Probe Library

Labelling

Single colour

Sample A

Array Manufacture

Hybridisation

Scanning

Array profiling

Affymetrix Array Targets

• Arabidopsis Genome 24,000

• C. elegans Genome 22,500

• Drosophila Genome 18, 500

• E. coli Genome 20, 366

• Human Genome U133 Plus 47,000

• Mouse Genome 39, 000

• Yeast Genome

• S.cerevisiae 5, 841

• S. pombe 5, 031

• Rat Genome 30, 000

• Zebrafish 14, 900

• Plasmodium / Anopheles

• P. faciparum 4,300

• A. gambiae 14,900

• Barley (25,500), Soybean (37,500 + 23,300 pathogen), Grape (15,700)

• Canine (21,700), Bovine (23,000)

• B.subtilis (5,000), S. aureus (3,300 ORFS), Xenopus (14, 400)

Closed System – Microarray

• Pros• High-throughput

• Targeted profiling

• Inexpensive – “population friendly”

• Analytical methods are standardised

• Negative• “Closed system” , novel = invisible

• Difficult to see allelle-specific expression

• Biases due to hybridisation• SNPs• Competitive and non-specific hybridisation

Open systems – RNA Sequencing

Technology:• Illumina• SOLiD, IonTorrent• 454

Pros:• Transcript discovery• Allelic expression• High resolution abundance measures

Cons:• Analysis can be complex• Expensive• Sensitivity is sequencing depth dependent

RNA Sequencing

Mortazavi et al., 2008

RNASeq - Correspondence

• Range > 5 orders of magnitude

• Better detection of low abundance transcripts

Marioni et al., 2009

Platform Choice / Sample Preparation Choice

What do you want to profile ?

• Polyadenylated• PolyA RNA extraction

• Small RNA (< 100 bases)• Size filtering by gel

• Strand-specific

• RNA – Protein Interactions• RNA Immunoprecipitation (IP)

RNASeq - Workflow

Library Construction

Sample

Total RNA

PolyA RNA

Small RNA

Sequencing

Base calling & QC

Mapping to Genome

Assembly to Contigs

Differential Expression

SNP detection

Transcript structure

Secondary structure

Targets or Products

Illumina RNASeq : TruSeq

Small RNA sequencing

Small RNA

smallRNA separation: PAGE

small RNA < 35bp

Strand - specificity

Using adaptors Using chemical modification

SMART : addition of C’s on 5’ end

Ligation : 3’ and 5’ adaptors added sequentially

Levin et al., 2010

dUTP : Addition and removal after selection

CSIRO. Nescent August 2011 - Measuring Expression Levin et al., 2010

Non-polyA methods

• Total RNA extraction

• Ribosomal RNA and tRNA > 95-97% of total RNA

• Ribosomal reduction methods• Subtractive hybridisation with rRNA probes

• Exonuclease cleave of rRNA

• NuGen – “proprietary combination of reverse transcriptase and primers in the Ovation RNA-Seq System”

• cDNA normalisation methods• Partial digestion of any highly abundant species (Evrogen)

Platform Choice / Sample Preparation Choice

What do you want to profile ?

• Polyadenylated• PolyA RNA extraction

• Small RNA (< 100 bases)• Size filtering by gel

• Strand-specific

• RNA – Protein Interactions• RNA Immunoprecipitation (IP)

• Non - PolyA• rRNA reduction

EXPERIMENTAL DESIGN and ANALYSIS

• Issues:• sequencing depth - how much ?

• number of replicates – how many ?

• Aims of the data : • Transcriptome assembly / transcript characterisation

• Maximise depth

• Detection of differential expression (denovo or reference)

• Balance depth and replication

CSIRO. Sequencing Depth V.S. Number of Replicates

RNASeq Experimental Design

Defining Replicates

• Technical Replicates • Biological Replicates

Library 1

Lane 1

Individual

Library 2

Lane 2 Lane 3 Lane 4 Lane 1

,Individual 1

Lane 2

Individual 2

Library 1 Library 2

Depth = 2 x 100% lane / sample 100% lane / sample

Lane 1

Library 4

Multiplex

Library 3

Library 2

Library 1

25% lane / sample

Coverage Depth

Number of Replicates

edgeR <= 0.01 , DESeq <= 0.01

More information in biological replicates than depth

For differential expression

2 4 6 8 10 12

False P

0.03 0.03 0.03 0.03 0.03 0.03

False N

0.84 0.72 0.64 0.59 0.54 0.50

True P

0.16 0.28 0.36 0.41 0.46 0.50

True N

0.97 0.97 0.97 0.97 0.97 0.97

RNASeq Analysis

• Overall Aim :• To get an accurate measurement of transcript abundance, structure

and identity

• Biases and Compositions

• Alignment• TopHat / Cufflinks

• Assembly• ABySS

Assumptions

Every transcript / k-mer has equal chance of being sequenced

No. sequences observed ≈ transcript abundance

Gene A = z Reads / million Gene B = y Reads / million

z = 2 x y

Gene A > Gene B

Length Bias

Oshlack and Wakefield, 2009

Alignment Bias

Sequencing Bias

Hansen et al., 2010

Gene A = z Reads / million / kb Gene B = y Reads / million / kb

Weighting schemas (e.g. Cufflinks) :

• Mapability

• kmer / fragment frequencies

Gene A1 = z Reads per million Gene A2 = y Reads per million

z = 2 x y

Sample A vs Sample B

Read density variability

RNASeq – Compositional properties

Depth of Sequence• Sequence count ≈ Transcript Abundance

• Majority of the data can be dominated by a small number of highly abundant transcripts

• Ability to observe transcripts of smaller abundance is dependent upon sequence depth

• Fixed budget of reads

A simple example – compositional bias

sample II

Sequencing budget / depth: 4000 reads

DDCCBB

sample IExpected counts

Expected counts

Soil diversity by phylogenetic analysis - Phylum level

Recognized bacterial phyla

0% 20% 40% 60% 80% 100%

% distribution

454-sequence analysis of bacterial 16S rRNA gene~410,000 sequences

A. Richardson, CSIRO

RNASeq Bioinformatics Analysis

• Aims:• To get an accurate measurement of transcript abundance,

structure and identity

• Biases and Compositions• Relative abundances NOT absolute

• Alignment• TopHat

• Assembly• ABySS

RNA Sequencing analysis

Sequence Data

Alignment

Read Density

Transcript Characterisation

Assembly

Contigs

Genome?

RNASeq – Alignment Considerations

Reads with multiple locations

• Discard / Random Allocation

• Clustering - local coverage

• Weighting

Reads Spanning Exons

• Make and align to exon junction libraries

• Denovo junction detection

Summarisation of counts

• Exons

• Transcript boundaries

• Inferred read boundaries

TopHat

Trapnell et al., 2009; Roberts et al., 2011

Multimapping : ≤10 sites

Assembly : consensus ‘island’ exon

TopHat / Cufflinks

Trapnell et al., 2009; Roberts et al., 2011

Heuristics :

• “Correct” errors in low coverage areas

• Grabs 45 bp either side of islands to capture splice sites

• Collapse small islands

• Looks for junctions within larger islands, highly covered

Cufflinks :

• calculates the probability of observing a certain fragment within a given transcript given surrounding fragments.

Alignment

• Great if you have a fully annotated, reference

• Okay.. If you have a partially annotated reference

• “Different” if you have a big bunch of ESTs

Options:• Align to a neighbouring genome or EST library• Denovo transcriptome assembly

Tools:• ABySS, Mira, Trinity, HT-Seq, SAMtools

RNA Sequencing analysis

Sequence Data

Alignment

Read Density

Transcript Characterisation

Assembly

Contigs

Genome?

Denovo transcriptome assembly

• ABySS• MIRA• Trinity• Velvet• AllPaths• Soap-denovo• Euler• CABOG• Edena• SHARCGS• VCAKE• SSAKE• CAP3

• Will run on reasonable computer resources for large genomes

• (e.g. < 1 TB of RAM)

• Paired end data handling

• Platform flexible

• Handles haplotype complexity and polyploid genomes

Denovo transcriptome assembly

• ABySS• MIRA• Trinity• Velvet• AllPaths• Soap-denovo• Euler• CABOG• Edena• SHARCGS• VCAKE• SSAKE• CAP3

• Will run on reasonable computer resources for large genomes

• (e.g. < 1 TB of RAM)

• Handles paired end data

• Handles data from all platforms

• Handles haplotype complexity and polyploid genomes

Assembly – Kmer graphs

Miller et al., 2010

• Sequencing error

Bubbles

• Polymorphism

Frayed Rope / Cycles

• Repeats

Miller et al., 2010

Bubbles

• Polymorphism

Frayed Rope / Cycles

• Repeats

Miller et al., 2010

ABySS & TransABySS

• User specifies k

• Optimal k depends on sequencing depth

ABySS & TransABySS

• Sequencing depth is relative to transcript abundance• Iterate over multiple k and merge

• Contigs contained within a large contig are “buried”

Assessing assembly quality ?

• Comparisons between assembly algorithms• Contig summary statistics• Comparisons to known resources (e.g. ESTs)

Trial on Rice Transcriptome:• 120 Million 75 bp single end Illumina reads – embryo

• ABySS :• Number of contigs = 6, 804• Contig length range = 38 – 2,818 [mean = 203]

• Database comparisons :

• Rice public cDNA sequences : 67, 393

• Contigs with high quality matches to cDNA : 6,555 (96%)

RNASeq Bioinformatics Analysis

• Aims:• To get an accurate measurement of transcript abundance,

structure and identity

• Biases and Compositions• Relative abundances NOT absolute

• Alignment

• Assembly

STATISTICAL ISSUES

Measuring Expression – Statistical Issues

• Data elements

• Normalisation

• Detection of Differential Expression

Count Data : of what ?

Garber et al., 2011

Statistical analysis of RNASeq

• Count data• Distribution is positively skewed, not normal• Between sample variability in counts - normalisation

Normalization is required

Two scenarios :

1. Different sizes of total reads (library size)

2. Fixed library size, subset of highly expressed reads in 1 sample.

Both reduce sequencing budget available for the majority of transcripts

Normalisation

• Assume the majority of log ratios = 0 [No change]

Robinson and Oshlack, 2010

TMM : Trimmed Mean of M values (log ratios)

Adjust TMM to be equal between samples

DE genes with and without TMM normalization

RNASeq data – Poisson Distributions

• Poisson distributions are used when things are counted

• The probability of seeing n events in a fixed time or space

• The number of lions on a 1 day safari

• The number of raindrops on a tennis court

• The number of flying elephants in a year

• Requires λ : rate of events• Variance = mean = λ

RNASeq data – Negative Binomial

• RNASeq data is more variable than Poisson• Variance > mean = λ

• Less prominent for large mean

• Over-dispersed Poisson

Noise types• Shot noise

• Unavoidable, prominent for low mean

• Technical noise• Small, hopefully, can be managed

• Biological noise• Sample differences

RNA Seq

• Variance also depends on the mean

Anders, 2010

RNASeq Model

The total counts for a transcript in sample j from condition c :

cjcj vss 2

Library normalisation

Mean Value Fitted Variance (overdispersion)

For a given gene , test for a difference in counts between conditions.

Is mean c1 + mean c2 statistically different to mean c1 + mean c1?

RNASeq DE Testing

• DESeq – Anders and Huber, 2010• EdgeR – Robinson et al., 2009 – R• BaySeq – Hardcastle and Kelley, 2010 – R• DEGSeq – Wang et al., 2010 – R• NBP - Di et al., 2011

• LOX – Zhang et al., 2010• Infers expression measures allowing for incorporation of noise from

different methodologies in the one experimental design

Measuring Expression• What & Why

• What is expression and why do we care?

• How• Platforms / Technology

• Closed approaches – Microarray• Open approaches - Sequencing

• Experimental Design

• Analysis• Biases• Bioinformatics• Statistical Issues and Analysis

• In action• Workshop – Detection of Differential Expression• Case Studies in Plant functional genomics

Contact UsPhone: 1300 363 400 or +61 3 9545 2176

Email: enquiries@csiro.au Web: www.csiro.au

Thank you

Plant IndustryJennifer M TaylorBionformatics Leader

Phone: +61 2 62464929Email: Jen.Taylor@csiro.au

Acknowledgements

Jose RoblesStuart StephenHua YingAndrew Spriggs

Alexie Pa

NESCENT Funding

NESCENT : NGS : Measuring expression Jen Taylor Bioinformatics Team CSIRO Plant Industry.

expression transcriptome

human genome sequencing

inherited genes

function polyadenylation

function availability

maria mcclafferty

abundance abundance

parts description function

Documents

CSIRO and OpenCL

TRAINER’S...

10/22/12 NGS Sequence data NGS Sequence...

CSIRO-Chile ICE International Centre of Excellence Mining...

What is NESCent? Inspired by the National Center for...

NESCent Postdoc Professional Development Series on Effective...

CSIRO ESG WORKSHOP

Meal plans - CSIRO

Evolution Education Resources from NESCent The National...

OpenTree at NESCent Academy 2012

Data licensing in CSIRO CSIRO INFORMATION MANAGEMENT &...

RNA-seq library prep introduction NESCent Academy.

preview - CSIRO Publishing

Sarah stephenson csiro

HyResource - CSIRO Research

CSIRO; Swinburne