![Page 1: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/1.jpg)
Detection and inference of differentially methylated regions from bisulfite
sequencing
Keegan Korthauer
@keegankorthauer
Bioconductor Meeting 2017
![Page 2: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/2.jpg)
Differential methylation
Differential methylation commonly studied in – Cancer – Developmental stages / Aging – Tissue types
Control
Disease
Unmethylated CpG
Methylated CpG
* * * *
Differentially Methylated Region (DMR)
![Page 3: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/3.jpg)
Whole Genome Bisulfite Sequencing
Treat DNA with bisulfite before sequencing – Methylated C -> unaffected – Unmethylated C -> appear as T
GAGCGATGGATAGCG CGAGTGATGGATAGC GTTACGAGCGACGG TGGTTACGAGCGATG
...TCTCGGTTACGAGCGACGGATAGCG...
Bisulfite sequencing reads Reference genome
![Page 4: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/4.jpg)
Whole Genome Bisulfite Sequencing
Treat DNA with bisulfite before sequencing – Methylated C -> unaffected – Unmethylated C -> appear as T
GAGCGATGGATAGCG CGAGTGATGGATAGC GTTACGAGCGACGG TGGTTACGAGCGATG
...TCTCGGTTACGAGCGACGGATAGCG...
Bisulfite sequencing reads Reference genome
3/3 1
3/4 0.75
1/4 0.25
0/1 0
Methylation Counts and Proportions
![Page 5: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/5.jpg)
Methods for DMR detection
Challenges – Have to handle small sample sizes ! – Accommodate known sources of variability – Detect region boundaries
Two main strategies – Detect significant CpGs, then group together neighbors to
form regions (DSS, BSmooth) – Targeted regions / sliding windows specified in advance
(BiSeq, MOABS)
![Page 6: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/6.jpg)
Genomicloca+on
Methyla+o
nDiffe
rence
0
Danger of grouping significant CpGs
![Page 7: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/7.jpg)
Genomicloca+on
Methyla+o
nDiffe
rence
0
TruePosi+veFalsePosi+veTrueNega+ve
Danger of grouping significant CpGs False Discovery Rate = # False Discoveries (FDR) Total # Discoveries = 2/8 = 0.25
![Page 8: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/8.jpg)
Genomicloca+on
Methyla+o
nDiffe
rence
0
Danger of grouping significant CpGs
TruePosi+veFalsePosi+veTrueNega+ve
False Discovery Rate = # False Discoveries (FDR) Total # Discoveries = 1/2 = 0.5
![Page 9: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/9.jpg)
Met
hyla
tion
0.2
0.5
0.8
●
●
Left VentricleSmall Intestine
chr5: 75,654,594 − 75,655,232 (width = 639), FDR: 0.0054, Stat: 23.4
Island Shore Shelf Open SeaCpGs
Exons
DMRs come in all shapes and sizes M
ethy
latio
n
0.2
0.5
0.8
●
●
Left VentricleSmall Intestine
chr16: 54,288,213 − 54,292,817 (width = 4,605), FDR: 0.0054, Stat: −23.678
Island Shore Shelf Open SeaCpGs
IRX3Exons
-> Detect region boundaries from the data
![Page 10: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/10.jpg)
Our approach: dmrseq 1. Detect regions by scanning the genome for candidate
regions
2. Region-level summary scores compared to a null to evaluate significance
Diffe
rence
0Diffe
rence
0
? ?
![Page 11: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/11.jpg)
Our approach: dmrseq 1. Detect regions by scanning the genome for candidate
regions
2. Region-level summary scores compared to a null to evaluate significance
Diffe
rence
0Diffe
rence
0
? ?
![Page 12: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/12.jpg)
Our approach: dmrseq 1. Detect regions by scanning the genome for candidate
regions
2. Region-level summary scores compared to a null to evaluate significance
Diffe
rence
0Diffe
rence
0
? ?
![Page 13: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/13.jpg)
Region level Summary Scores
– Biological variability
among samples
– Correlation among nearby CpGs
– Higher variability in CpGs with lower coverage
Generalized Least Squares (GLS) regression
– Nested autoregressive correlated error structure
– Coverage-weighted variance
arcsin(2π ijr −1) = αlr1[i=1]l∑ +βrX j +εijr
Var(εijr ) =
Summary statistic that is approximately exchangeable across the genome so we can
generate a pooled null
![Page 14: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/14.jpg)
dmrseq accurately controls FDR
0.0
0.1
0.2
0.3
0.4
0.0 0.1 0.2 0.3 0.4Specified FDR level
Obse
rved
FDR
leve
l
(A) dmrseq
0.0
0.1
0.2
0.3
0.4
0.0 0.1 0.2 0.3 0.4Specified FDR level
(B) metilene
SimulationD2
D3
![Page 15: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/15.jpg)
dmrseq has high sensitivity
0.00
0.25
0.50
0.75
1.00
0.00 0.25 0.50 0.75 1.00FDR
Powe
r
(A) Simulation D2
0.00
0.25
0.50
0.75
1.00
0.00 0.25 0.50 0.75 1.00FDR
(B) Simulation D3
MethodBSmooth
dmrseq
DSS
metilene
![Page 16: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/16.jpg)
DMRs enriched for associations with differential expression
(A) Roadmap Tissue Comparisons
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
Right Ventricle vs Small Intestine
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Right Ventricle vs Sigmoid Colon
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Left Ventricle vs Small Intestine
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
Left Ventricle vs Sigmoid Colon
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Sigmoid Colon vs Small Intestine
(B) Murine Leukemia Models
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
AML vs Control
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
ALL vs Control
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
AML vs ALL
Method BSmooth dmrseq DSS metilene
(A) Roadmap Tissue Comparisons
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
Right Ventricle vs Small Intestine
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Right Ventricle vs Sigmoid Colon
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Left Ventricle vs Small Intestine
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
Left Ventricle vs Sigmoid Colon
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Sigmoid Colon vs Small Intestine
(B) Murine Leukemia Models
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
AML vs Control
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
ALL vs Control
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
AML vs ALL
Method BSmooth dmrseq DSS metilene
(A) Roadmap Tissue Comparisons
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
Right Ventricle vs Small Intestine
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Right Ventricle vs Sigmoid Colon
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Left Ventricle vs Small Intestine
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
Left Ventricle vs Sigmoid Colon
1
2
3
4
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
Sigmoid Colon vs Small Intestine
(B) Murine Leukemia Models
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
log2
odd
s
AML vs Control
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
ALL vs Control
0
1
2
3
0.01 0.10 0.25 0.50 0.75 1.00FDR threshold (square root scaled)
AML vs ALL
Method BSmooth dmrseq DSS metilene
![Page 17: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/17.jpg)
dmrseq Summary
- dmrseq identifies and prioritizes DMRs from bisulfite sequencing experiments
- Computes region summary statistics that account for known sources of variability across the genome
- Achieves accurate False Discovery Rate control by generating a null distribution that pools information across the genome
![Page 18: Detection and inference of differentially methylated regions from … · 2017-08-08 · Keegan Korthauer @keegankorthauer Bioconductor Meeting 2017 . Differential methylation Differential](https://reader034.vdocuments.site/reader034/viewer/2022050506/5f97f666bbb53c04170ce917/html5/thumbnails/18.jpg)
Rafa Lab Rafael Irizarry Chinmay Shukla
Collaborators Sutirtha Chakraborty Yuval Benjamini
Learn More Slides goo.gl/MwQz5f dmrseq available on GitHub https://github.com/kdkorthauer/dmrseq
Acknowledgements
Contact [email protected] @keegsdur