Computational methods for region-based analysis of epigenome signals

Nathan Sheffield, PhD
www.databio.org/slides

outline

The (epi)genome revolution
Epigenome tools
|
|

20%
40%
40%
|

Project organization
◁ Questions ▷

The genome revolution


A revolution driven by DNA sequencing technology
Sequencing technology can also measure epigenome signals

Epigenomics is the study of the chemical modification and physical conformation of cellular DNA and bound proteins

Rosa et al. 2013
  • Histone modification:
    ChIP-seq
  • DNA methylation:
    Bisulfite-seq
  • Chromatin accessibility:
    ATAC-seq

The Sequence Read Archive is growing

https://www.ncbi.nlm.nih.gov/sra/docs/sragrowth/
Data is becoming more...

abundant
available
powerful
So why are the world's problems not solved?


First step in bioinformatics analysis:

pipeline

Papers with
"bioinformatics pipeline"
in title
Problem solved?
Problem solved?
Problem solved?

Data munging

Then, downstream tools need a different organization

What if?

Microwave syndrome

In user interface design, prioritizing easy access to integrated functions over their individual components.

The UNIX philosophy

[T]he power of a system comes more from the relationships among programs than from the programs themselves.

Many UNIX programs do quite trivial tasks in isolation, but, combined with other programs, become general and useful tools.

- Kernighan and Pike, The UNIX Programming Environment (1983, p. viii)

Problem

Solution

Problem

Solution

PEP: Portable Encapsulated Projects

PEP format

project_config.yaml 
metadata:
  sample_annotation: /path/to/samples.csv
  output_dir: /path/to/output/folder
samples.csv 
sample_name, protocol, organism, data_source
frog_0h, RNA-seq, frog, /path/to/frog0.gz
frog_1h, RNA-seq, frog, /path/to/frog1.gz
frog_2h, RNA-seq, frog, /path/to/frog2.gz
frog_3h, RNA-seq, frog, /path/to/frog3.gz

PEP portability features

Derived attributes
Implied attributes
Subprojects
Derived attributes
Automatically build new sample attributes from existing attributes.
Without derived attribute:
| sample_name | t | protocol | organism | data_source | | ------------- | ---- | :-------------: | -------- | ---------------------- | | frog_0h | 0 | RNA-seq | frog | /path/to/frog0.gz | | frog_1h | 1 | RNA-seq | frog | /path/to/frog1.gz | | frog_2h | 2 | RNA-seq | frog | /path/to/frog2.gz | | frog_3h | 3 | RNA-seq | frog | /path/to/frog3.gz |
Using derived attribute:
| sample_name | t | protocol | organism | data_source | | ------------- | ---- | :-------------: | -------- | ---------------------- | | frog_0h | 0 | RNA-seq | frog | my_samples | | frog_1h | 1 | RNA-seq | frog | my_samples | | frog_2h | 2 | RNA-seq | frog | my_samples | | frog_3h | 3 | RNA-seq | frog | my_samples | | crab_0h | 0 | RNA-seq | crab | your_samples | | crab_3h | 3 | RNA-seq | crab | your_samples |
| sample_name | t | protocol | organism | data_source | | ------------- | ---- | :-------------: | -------- | ---------------------- | | frog_0h | 0 | RNA-seq | frog | my_samples | | frog_1h | 1 | RNA-seq | frog | my_samples | | frog_2h | 2 | RNA-seq | frog | my_samples | | frog_3h | 3 | RNA-seq | frog | my_samples | | crab_0h | 0 | RNA-seq | crab | your_samples | | crab_3h | 3 | RNA-seq | crab | your_samples |
Project config file: 
derived_columns: [data_source]
data_sources:
  my_samples: "/path/to/my/samples/{organism}_{t}h.gz"
  your_samples: "/path/to/your/samples/{organism}_{t}h.gz"
{variable} identifies sample annotation columns
Benefit: Enables distributed files, portability
Implied attributes
Add new sample attributes conditioned on values of existing attributes
Before:
| sample_name | protocol | organism | | ------------- | :-------------: | -------- | | human_1 | RNA-seq | human | | human_2 | RNA-seq | human | | human_3 | RNA-seq | human | | mouse_1 | RNA-seq | mouse |
After:
| sample_name | protocol | organism | genome | | ------------- | :-------------: | -------- | ------ | | human_1 | RNA-seq | human | hg38 | | human_2 | RNA-seq | human | hg38 | | human_3 | RNA-seq | human | hg38 | | mouse_1 | RNA-seq | mouse | mm10 |
| sample_name | protocol | organism | | ------------- | :-------------: | -------- | | human_1 | RNA-seq | human | | human_2 | RNA-seq | human | | human_3 | RNA-seq | human | | mouse_1 | RNA-seq | mouse |
Project config file: 
implied_columns:
  organism:
    human:
      genome: hg38
    mouse:
      genome: mm10
Benefit: Divides project from sample metadata
Subprojects
Define activatable project attributes.
subprojects:
  diverse:
    metadata:
      sample_annotation: psa_rrbs_diverse.csv
  cancer:
    metadata:
      sample_annotation: psa_rrbs_intracancer.csv
Benefit: Defines multiple similar projects in a single file
How is this portable and encapsulated?

Encapsulated:
A project is an extensible object, with samples and settings as attributes.
Portable:
  1. A project can be moved from one analysis tool to another
  2. A project can be moved from one computing environment to another

Locus Overlap Analysis

http://code.databio.org/LOLA/
Sheffield and Bock (2016). Bioinformatics.
Nagraj, Magee, and Sheffield (2018). Nucleic Acids Research.

If subject list has no containment,
identifying overlaps is fast

binary search on start intervals, followed by backward steps:

The problem arises with contained interval overlaps

How can we improve efficiency
without guaranteeing no containment?

Many approaches to solve the 'containment' issue:

- Nested Containment Lists (GRanges) [@Alekseyenko2007; @Aboyoun2012] - R-trees (bedtools) [@Kent2002; @Quinlan2010], Augmented interval trees [@Cormen2001] These methods try to structure the data to provide non-containment guarantees

Methods provide non-containment guarantees

R-trees

Annotates tree nodes with a minimum bounding rectangle of elements. A query that does not intersect the bounding rectangle will not intersect any child element.

Nested Containment Lists

Augmented Interval List

1. Augment the list with the running maximum *end* value. *solves the problem for lowly-contained lists* 2. Decompose the list to minimize containment. *extends the solution to highly-contained lists*

Augment with the running maximum end value, `maxE`

Provides a local guarantee of no containment.

AIList works on contained lists

But long containment runs are problematic

Decompose long runs with constant `maxE`

Performance

  • How does the `maxE` minimum run length affect performance?
  • How does it compare to existing approaches?
  • How does it scale with increasing size of subject?

Datasets

How does the `maxE` minimum run length affect performance?

How does it compare to existing approaches?

How does it scale with increasing size of subject?

Conclusion

  • Augmented Interval Lists add the maximum running end value to a list of intervals
  • The data structure is simpler than other methods
  • AILists improve performance, particularly in highly contained interval sets

Conclusion

Pepkit provides a start-to-finish toolkit for processing epigenome data.

pepkit.github.io
LOLA is one of our tools to ask biological questions of genomic regions

code.databio.org/LOLA

Thank You


Sheffield lab
John Lawson
Vince Reuter
Jason Smith
Jianglin Feng
Michal Stolarczyk
Aaron Gu
Ognen Duzlevski


SOM research computing
Pete Nagraj
Neal Magee
Funding:




nsheff · databio.org · nsheffield@virginia.edu