1. Al Ritacco, Shailender Nagpal Research Computing
RNA-seq for Transcriptome profiling and discovery of novel transcripts and alternatively spliced variants using HPC
Presented by:
Al Ritacco, Shailender Nagpal Research Computing
UMASS Medical School
Information Services,
09/17/2012

2. Agenda SESSION 1 SESSION 2 What is RNA-seq?
Workflow for RNA-seq analysis
Tools required
SESSION 2
Download and perform QC on sample dataset
Mapping and alignment
Transcript expression and other “discoveries”
Information Services,
00/00/2010

3. Agenda SESSION 1 SESSION 2 What is RNA-seq?
Workflow for RNA-seq analysis
Tools required
SESSION 2
Download and perform QC on sample dataset
Mapping and alignment
Transcript expression and other “discoveries”
Information Services,
00/00/2010

4. What is “Next Generation Sequencing”?
Set of new high throughput technologies
allow millions of short DNA sequences from a biological sample to be “read” or sequenced in a rapid manner
Computational power is then used to assemble or align the “reads” to a reference genome, allowing biologists to make comparisons and interpret various biological phenomena
Due to high depth of coverage (30-100x), accurate sequencing is obtained much faster and cheaper compared to traditional Sanger/Shotgun sequencing

5. RNA-Seq experiment
Reverse transcription of mRNAs yield double stranded cDNAs, which are sliced to selected fragment length

6. What is RNA-Seq?
RNA-seq is a Next Generation Sequencing (NGS) technology for sequencing total mRNA (“expressed”) in biological samples of interest such as tissues, tumors and cell lines
Provides deep coverage and base level resolution
Abundance of known transcripts
Novel transcripts
Alternative splicing events
Post-transcriptional mutations
Gene fusions

7. RNA-seq reads: Issue 1
ACTTAAGGCTGACTAGC
TCGTACCGATATGCTG
Small, single-end reads are hard to align to a reference genome - multiple possible mapping sites. Longer reads can overcome this limitation
Paired ends allow for longer fragments to be sequenced, with a small read from each end of the fragment. Distance between ends validates mapping

8. RNA-seq reads: Issue 2 Reads from one exon are easily mapped
Reads that span 2 exons are handled by special software like TopHat
Information Services,
00/00/2010

9. DNA assembly versus alignment
DNA assembly is the computational task of putting together pieces of the genome in the original order
Overlapping short sequences are extended to form islands, that are subsequently extended and merged
Used for denovo sequencing of a new genome
DNA alignment is the computational task of mapping short reads to a known, sequenced genome
Reads and searched for in the full genome sequence, then aligned with a local alignment algorithm

10. Agenda SESSION 1 SESSION 2 What is RNA-seq?
Workflow for RNA-seq analysis
Tools required
SESSION 2
Download and perform QC on sample dataset
Mapping and alignment
Transcript expression and other “discoveries”
Information Services,
00/00/2010

11. RNA-seq workflow summary
Sample preparation and submission for experiment. Obtain “read” file(s) as FASTQ
Perform QC
Perform read alignment and mapping to reference genome
Determine expression, novel transcripts and alternative splicing

12. Step 1: Working with FASTQ Reads
After a sample has been processed by an NGS platform, DNA sequence “reads” are provided to the user in FASTQ format
FASTQ is a text-based format for storing both a DNA sequence and its corresponding quality scores 
sequence letter and quality score are encoded with a single ASCII character for brevity
originally developed at the Wellcome Trust Sanger Institute to bundle a FASTA sequence and its quality data
recently become the de facto standard for storing the output of high throughput sequencing instruments

13. Type of reads Single end reads Paired end reads
refer to the sequence determined as DNA bases are added to single stranded DNA and detected, usually from one end only
Paired end reads
refer to the two ends of the same DNA molecule
After sequencing one end, you can turn it around and sequence the other end.
Long segment of DNA in between the two ends (usually bp), who’s sequence is unknown
Once the two paired end reads are mapped, the intermediate sequence can be inferred from reference sequence

14. Quality Score
A quality value Q is an integer mapping of p (i.e., the probability that the corresponding base call is incorrect)
Two different equations have been in use. The first is the standard Sanger variant to assess reliability of a base call, otherwise known as Phred quality score:
The Solexa pipeline earlier used a different mapping, encoding the odds p/(1-p) instead of the probability p:
Although both mappings are asymptotically identical at higher quality values, they differ at lower quality levels (i.e., approximately p > 0.05, or equivalently, Q < 13)

15. Quality Score Encoding
Various platform produce versions of FASTQ format, which mainly differ in the Quality score representation
Converters from various tools can convert Solexa to Sanger FASTQ format

16. Step 2: Quality Control
FASTQ, PRINSEQ or other custom tools can be used to perform QC on FASTQ files
Good tutorial for FASTQC on Youtube:
Information Services,
00/00/2010

17. Step 3: Alignment/Mapping to Reference Genome
Many tools exist that will map the reads to the reference genome and align them, generating a quality score per base of alignment
BLAST-like algorithm to search the read in the genome
Local alignment to determine the accuracy of the alignment
Many tools (SAMtools, MAQ, TopHat, CASSAVA, etc) use the BWA, Bowtie and Eland algorithms.Result of alignment is the SAM file

18. Alignment algorithms
A category of aligners that hash the reads and scan the genome for matches
Eland, RMAP, MAQ, ZOOM, SeqMap, CloudBurst, SHRiMP
Disadvantages
For few reads, whole genome must be scanned
Memory footprint is variable

19. Alignment algorithms (…contd)
Another category of aligners hash the reference genome
SOAPv1, PASS, MOM, ProbeMatch, NovoAlign, ReSeq, Mosaik, Bfast
Easily parallelized with multi-threading, but they usually require large memory to build an index for the human genome
Disadvantage: iterative strategy frequently introduced by these software may make their speed sensitive to the sequencing error rate

20. Alignment algorithms (…contd)
A third category which does alignment by merge-sorting the reference subsequences and read sequences
Slider
Burrows–Wheeler Transform (BWT) for string matching has been incorporated into a new generation of alignment tools
SOAPv2, Bowtie and BWA
These are memory efficient and suit well to single and paired-end read alignments

21. Step 4: Transcript expression
“Cufflinks”, “myRNA” and other software can estimate the expression level/ abundance of the transcripts
Other discoveries possible – novel transcripts, fusions, etc.

22. Agenda SESSION 1 SESSION 2 What is RNA-seq?
Workflow for RNA-seq analysis
Tools required
SESSION 2
Download and perform QC on sample dataset
Mapping and alignment
Transcript expression and other “discoveries”
Information Services,
00/00/2010

23. FASTQ manipulation tools
Galaxy FASTQ tools
FASTXtoolkit

24. QC software
FASTqc
PRINSEQ
Information Services,
00/00/2010

25. Alignment software
Single and paired-end alignments can be done using the BWA algorithm in the following software
MAQ (
BWA
Bowtie
TopHat (
Alignments are produced generally in the widely accepted SAM format

26. Step 4: Transcript expression
Cufflinks
myRNA

27. Agenda SESSION 1 SESSION 2 What is RNA-seq?
Workflow for RNA-seq analysis
Tools required
SESSION 2
Download and perform QC on sample dataset
Mapping and alignment
Transcript expression and other “discoveries”
Information Services,
00/00/2010

28. Computing Hardware Requirements
Two types of computing hardware are ideally suited for NGS data analysis
High-end workstation, for example: 64-bit linux, 3.6 GHz quad-core processor, 32 GB RAM, 7200 rpm hard disk
High Performance Computing cluster (HPC) where total execution time can be sped up – for example, split reads into small files and align them in parallel on dozens of nodes
For this workshop, we will use HPC

29. RNA-seq datasets
NCBI’s Short Read Archive is a good source of datasets. Data can be download from:
Look for HapMap, 1000 genomes, cancer samples – all types of experiment types and platforms
Combine keywords in search: “Illumina” and “Paired” and “RNA-seq” and “HapMap”

30. Obtain data from SRA Dataset to be used:
Prostate cancer
Tumor-matched Normal
These are two paired-end Solexa datasets with reads split by those belong to the forward or end of a sequence fragment

31. Setting up the data for analysis
Create directory for this dataset
cd /home/username
mkdir rna-seq
cd rna-seq
Information Services,
00/00/2010

32. Load RNA-seq tools Load TopHat, Bowtie, Cufflinks and Samtools
module load tophat-1.2.0
module load cufflinks-2.0.2
module load bowtie
module load samtools
module load sratoolkit.2.1.9

33. Download Genome Annotation (GTF) file
The genome annotation file contains exon/intron co-ordinates of reference genome

34. Download the dataset
Download the Paired-end reads for the following library
wget ftp://ftp-trace.ncbi.nlm.nih.gov/sra/sra-instant/reads/ByRun/sra/SRR/SRR057/SRR057652/SRR sra
wget ftp://ftp-trace.ncbi.nlm.nih.gov/sra/sra-instant/reads/ByRun/sra/SRR/SRR057/SRR057634/SRR sra

35. SRA tools to extract FASTQ files
Use "sratools" to convert SRA format to FASTQ
fastq-dump -A SRR split-3 SRR sra
fastq-dump -A SRR split-3 SRR sra
Information Services,
00/00/2010

36. Perform QC using FASTQC
Load FASTQC module
module load fastqc-1.0
Perform QC
fastqc –t 8 SRR057634_1.fastq
fastqc –t 8 SRR057634_2.fastq
This creates 2 directories with output in HTML format for visual inspection in a browser, key statistics and tests are in text file
The zip files generated can be deleted

37. Agenda SESSION 1 SESSION 2 What is RNA-seq?
Workflow for RNA-seq analysis
Tools required
SESSION 2
Download and perform QC on sample dataset
Mapping and alignment
Transcript expression and other “discoveries”
Information Services,
00/00/2010

38. TopHat-Cufflinks
Information Services,
00/00/2010

39. Alignment and Mapping
For RNA-seq, the reads will either map fully to exons or partially to exon-intron junctions, resulting in rejected reads for the latter case
Should not use Bowtie or BWA directly for mapping against a reference genome
one of the goals is to identify novel transcripts, so we should not use transcriptome as reference
TopHat is ideally suited for the job, uses Bowtie for read alignment
needs the reference genome and it's annotations as an input. Annotations are optional

40. Running TopHat for read alignment
To run TopHat, execute the following command:
tophat --num-threads 8 \
--solexa-quals --max-multihits 10 \
--coverage-search --microexon-search \
--mate-inner-dist 150 \
-o tophat-tumor-out \
--keep-tmp \
–G /usr/public_data/ucsc/genomes/hg19/hg19.gtf \ hg19 \
SRR057634_1.fastq \
SRR057634_2.fastq

41. Working with TopHat output
This produces the “tophat-tumor-out” folder with the following files:
accepted_hits.bam
junctions.bed
insertions.bed
deletions.bed

42. Agenda SESSION 1 SESSION 2 What is RNA-seq?
Workflow for RNA-seq analysis
Tools required
SESSION 2
Download and perform QC on sample dataset
Mapping and alignment
Transcript expression and other “discoveries”
Information Services,
00/00/2010

43. Reporting quantitative expression: FPKM/RPKM
In NGS RNA-seq experiments, quantitative gene expression data is normalized for total gene/transcript length and the number of sequencing reads, and reported as
RPKM: Reads Per Kilobase of exon per Million mapped reads. Used for reporting data based on single-end reads
FPKM: Fragments Per Kilobase of exon per Million fragments. Used for reporting data based on paired-end fragments

44. Cufflinks to estimate expression
Quantify reference genes and transcripts only
cufflinks -p 8 -G hg19.gtf -o cuff-tumor accepted_hits.bam
Quantify novel genes & transcripts use hg19 as "guide”
cufflinks -p 8 -g hg19.gtf -o cuff-tumor accepted_hits.bam
Quantify novel genes & transcripts, "unguided"
cufflinks -p 8 -o cuff-tumor accepted_hits.bam

45. Cufflinks (…contd) This creates the following files
transcripts.gtf (Generated annotation)
isoforms.fpkm_tracking (Transcript expression)
genes.fpkm_tracking (Gene expression)
skipped.gtf (Skipped annotations)
Depending on the mode in previous step, these files can have vague identifiers (CUFF*) for gene names if known gene annotations are not used. We have to compare with reference annotations to uncover which genes they are
"Cuffcompare" allows us to do that

46. Compare assembled transcripts to reference
# Run “cuffcompare” for reference-guided assembly
cd cuff2
cuffcompare -r ../../hg19.gtf –R -V transcripts.gtf
# This produces the following files in the “cuffcom-out” directory with the following files
cuffcmp.transcripts.gtf.tmap,
cuffcmp.transcripts.gtf.refmap, cuffcmp.tracking, cuffcmp.stats, cuffcmp.loci, cuffcmp.combined.gtf

47. Cuffcompare output cuffcmp.stats cuffcmp.combined.gtf
Reports statistics related to the "accuracy" of the transcripts when compared to the reference annotation data.
Gene finding measures of “sensitivity” and “specificity” are calculated at various levels (nucleotide, exon, intron, transcript, gene)
cuffcmp.combined.gtf
Reports a GTF file containing the "union" of all transfrags in each sample

48. Cuffcompare output cuffcmp.loci cuffcmp.tracking
Each row contains a transcript structure that is present in one or more input GTF files
Column number
Column name
Example
Description 1
Cufflinks transfrag id
TCONS_
A unique internal id for the transfrag 2
Cufflinks locus id
XLOC_000023
A unique internal id for the locus 3
Reference gene id
Tcea
The gene_name attribute of the reference GTF record for this transcript, or '-' if no reference transcript overlaps this Cufflinks transcript 4
Reference transcript id
uc007afj.1
The transcript_id attribute of the reference GTF record for this transcript, or '-' if no reference transcript overlaps this Cufflinks transcript 5
Class code c
The type of match between the Cufflinks transcripts in column 6 and the reference transcript. See class codes

49. Cuffcompare output cuffcmp.transcripts.gtf.refmap
For each input GTF file, it lists the reference transcripts, one row per reference transcript for cufflinks transcripts that either fully or partially match it
Column number
Column name
Example
Description 1
Reference gene name
uc007crl.1
The gene_name attribute of the reference GTF record for this transcript, if present. Otherwise gene_id is used. 2
Reference transcript id
The transcript_id attribute of the reference GTF record for this transcript 3
Class code c
The type of match between the Cufflinks transcripts in column 4 and the reference transcript. One of either 'c' for partial match, or '=' for full match. 4
Cufflinks matches
CUFF ,CUFF
A comma separated list of Cufflinks transcript ids matching the reference transcript

50. Cufflinks output (…contd)
cuffcmp.transcripts.gtf.tmap
For each input GTF file, it lists the most closely matching reference transcript, one row per cufflinks transcript, for each cufflinks transcript
(see column definitions on next slide)

51. Cufflinks output (…contd)
Col.
Column name
Example
Description 1
Reference gene name
Myog
The gene_name attribute of the reference GTF record for this transcript, if present. Otherwise gene_id is used. 2
Reference transcript id
uc007crl.1
The transcript_id attribute of the reference GTF record for this transcript 3
Class code c
The type of relationship between the Cufflinks transcripts in column 4 and the reference transcript (see Class Codes) 4
Cufflinks gene id
CUFF.23567
The Cufflinks internal gene id 5
Cufflinks transcript id
CUFF
The Cufflinks internal transcript id 6
Fraction of major isoform (FMI)
100
The expression of this transcript expressed as a fraction of the major isoform for the gene. Ranges from 1 to 100. 7
FPKM
1.4567
The expression of this transcript expressed in FPKM 8
FPKM_conf_lo
0.7778
The lower limit of the 95% FPKM confidence interval 9
FPKM_conf_hi
1.9776
The upper limit of the 95% FPKM confidence interval 10
Coverage
3.2687
The estimated average depth of read coverage across the transcript. 11
Length
1426
The length of the transcript 12
Major isoform ID
The Cufflinks ID of the gene's major isoform

52. Cuffcompare output (…contd)
Class codes
Priority
Code
Description 1 =
Complete match of intron chain 2 c
Contained 3 j
Potentially novel isoform (fragment): at least one splice junction is shared with a reference transcript 4 e
Single exon transfrag overlapping a reference exon and at least 10 bp of a reference intron, indicating a possible pre-mRNA fragment. 5 i
A transfrag falling entirely within a reference intron 6 o
Generic exonic overlap with a reference transcript 7 p
Possible polymerase run-on fragment (within 2Kbases of a reference transcript) 8 r
Repeat. Currently determined by looking at the soft-masked reference sequence and applied to transcripts where at least 50% of the bases are lower case 9 u
Unknown, intergenic transcript 10 x
Exonic overlap with reference on the opposite strand 11 s
An intron of the transfrag overlaps a reference intron on the opposite strand (likely due to read mapping errors) 12 .
(.tracking file only, indicates multiple classifications)

53. Further analysis
The slides so far show how to interpret the RNA-seq results of one sample/library
In an actual experiment, tumor-normal pair might be available as separate library. Refer to tophat-cufflinks diagram to choose workflow
Might involve cuffmerge prior to cuffcompare
Cuffdiff and cuffquant can be used to report expression or even perform differential expression analysis
Information Services,
00/00/2010

54. Q&A
Don’t be shy – everyone has a different work flow and we want to help you
Information Services,
00/00/2010