Estimation of alternative
splicing isoform frequencies
from RNA-Seq data
Marius Nicolae
Computer Science and Engineering Department
University of Connecticut
Joint work with Serghei Mangul, Ion Mandoiu and Alex Zelikovsky
Outline
•
•
•
•
Introduction
EM Algorithm
Experimental results
Conclusions and future work
Alternative Splicing
[Griffith and Marra 07]
RNA-Seq
Make cDNA & shatter into fragments
Sequence fragment ends
Map reads
A
Gene Expression (GE)
B
C
Isoform Expression (IE)
D
E
Isoform Discovery (ID)
A
B
A
C
D
E
C
Gene Expression Challenges
• Read ambiguity (multireads)
A
B
C
• What is the gene length?
D
E
Previous approaches to GE
• Ignore multireads
• [Mortazavi et al. 08]
– Fractionally allocate multireads based on unique read
estimates
• [Pasaniuc et al. 10]
– EM algorithm for solving ambiguities
• Gene length: sum of lengths of exons that appear
in at least one isoform
 Underestimates expression levels for genes with 2 or
more isoforms [Trapnell et al. 10]
Read Ambiguity in IE
A
A
B
C
C
D
E
Previous approaches to IE
• [Jiang&Wong 09]
– Poisson model + importance sampling, single reads
• [Richard et al. 10]
— EM Algorithm based on Poisson model, single reads in exons
• [Li et al. 10]
– EM Algorithm, single reads
• [Feng et al. 10]
– Convex quadratic program, pairs used only for ID
• [Trapnell et al. 10]
– Extends Jiang’s model to paired reads
– Fragment length distribution
Our contribution
• EM Algorithm for IE
– Single and/or paired reads
– Fragment length distribution
– Strand information
– Base quality scores
– Hexamer bias correction
– Annotated repeats correction
Read-Isoform Compatibility
wr ,i
wr ,i   OaQa Fa
a
Fragment length distribution
• Paired reads
Fa(i)
i
A
B
j
A
C
C
Fa (j)
Fragment length distribution
• Single reads
i
A
B
j
A
C
C
Fa(i)
Fa (j)
IsoEM algorithm
E-step
M-step
Speed improvements
• Collapse identical reads into read classes
(i3,i5)
(i3,i4)
(i1,i2)
Reads
LCA(i3,i4)
i1
i2
i3
i4
i5
i6
Isoforms
Speed improvements
• Collapse identical reads into read classes
i2
i1
i4
i3
i5
i6
Isoforms
• Run EM on connected components, in parallel
Simulation setup
25000
100000
20000
10000
Number of genes
Number of isoforms
• Human genome UCSC known isoforms
15000
10000
5000
1000
100
0
10
1
10
100
1000
10000
100000
0
5
Isoform length
10 15 20 25 30 35 40 45 50 55
Number of isoforms
• GNFAtlas2 gene expression levels
– Uniform/geometric expression of gene isoforms
• Normally distributed fragment lengths
– Mean 250, std. dev. 25
Accuracy measures
• Error Fraction (EFt)
– Percentage of isoforms (or genes) with relative
error larger than given threshold t
• Median Percent Error (MPE)
– Threshold t for which EF is 50%
• r2
Error Fraction Curves - Isoforms
• 30M single reads of length 25
100
% of isoforms over threshold
90
80
Uniq
70
Rescue
60
UniqLN
50
Cufflinks
40
30
RSEM
20
IsoEM
10
0
0
0.2
0.4
0.6
Relative error threshold
0.8
1
Error Fraction Curves - Genes
• 30M single reads of length 25
100
% of genes over threshold
90
80
Uniq
70
Rescue
60
GeneEM
50
Cufflinks
40
RSEM
30
IsoEM
20
10
0
0
0.2
0.4
0.6
Relative error threshold
0.8
1
MPE and EF15 by Gene Frequency
• 30M single reads of length 25
Read Length Effect
• Fixed sequencing throughput (750Mb)
0.978
25
0.976
20
Median Percent Error
0.974
r2
0.972
0.97
0.968
0.966
Paired reads
0.964
Single reads
15
10
5
Paired reads
Single reads
0.962
0
25
35
45
55
65
Read length
75
85
95
25
35
45
55
65
Read length
75
85
95
Effect of Pairs & Strand Information
• 1-60M 75bp reads
0.985
0.98
0.975
r2
0.97
0.965
RandomStrand-Pairs
0.96
CodingStrand-pairs
0.955
RandomStrand-Single
CodingStrand-single
0.95
0.945
0
10,000,000
20,000,000
30,000,000
# reads
40,000,000
50,000,000
60,000,000
Validation on Human RNA-Seq Data
• ≈8 million 27bp reads from two cell lines [Sultan et al. 10]
• 47 genes measured by qPCR [Richard et al. 10]
IsoEM
3
3
2.5
2.5
Log(IsoEM)
Log(Cufflinks)
Cufflinks
2
2
1.5
1.5
R² = 0.5505
R² = 0.4604
1
1
1
1.5
2
2.5
Log(qPCR)
3
3.5
4
1
1.5
2
2.5
Log(qPCR)
3
3.5
4
Runtime scalability
• Scalability experiments conducted on a Dell PowerEdge R900
– Four 6-core E7450Xeon processors at 2.4Ghz, 128Gb of internal memory
160
140
RandomStrand-Pairs
120
CodingStrand-Pairs
CPU Seconds
RandomStrand-Single
100
CodingStrand-Single
80
60
40
20
0
0
Million
Fragments
10
20
30
Conclusions & Future Work
• Presented EM algorithm for estimating isoform/gene
expression levels
– Integrates fragment length distribution, base qualities, pair
and strand info
– Java implementation available at
http://dna.engr.uconn.edu/software/IsoEM/
• Ongoing work
– Comparison of RNA-Seq with DGE
– Isoform discovery
– Reconstruction & frequency estimation for virus quasispecies
Acknowledgments

NSF awards 0546457 & 0916948 to IM and 0916401 to AZ