Detection and inference of differentially methylated regions from bisulfite

sequencing

Keegan Korthauer

@keegankorthauer

Bioconductor Meeting 2017

Differential methylation

Differential methylation commonly studied in –  Cancer –  Developmental stages / Aging –  Tissue types

Control

Disease

Unmethylated CpG

Methylated CpG

* * * *

Differentially Methylated Region (DMR)

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

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

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)

Genomicloca+on

Methyla+o

nDiffe

rence

0

Danger of grouping significant CpGs

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

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

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

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

? ?

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

? ?

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

? ?

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

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

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

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

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

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 keegan@jimmy.harvard.edu @keegsdur