Day 5-2

1
Day 5-2
What bioinformatics tools can be used for
analysing ChIP data?
What bioinformatics tools can be used for
analysing ChIP data?
2
After this seminar

• You should be able to
• Understand the differences between CHip-chip and
  CHip-Seq and identify key decision making steps
  for choosing a platform
• Identify bioinformatics steps needed for handling
  CHip-chip and Chip-Seq datasets
• Understand underlying data from genome tiling
  arrays
• Understand how to search for binding sites in
  genomic data
• Understand the need for skills in handling large
  datasets

3
General problem

• Find accessible regions of DNA that are bound to
  your protein.
• What method is best?
• What sort of bioinformatics skills are required?
• What is real signal and what is noise?
• What do we do with the regions once you have
  identified them?

Zheng, M. et al. (2007) ChIP-chip data, model,
and analysis. Biometrics, Vol 63, 787-796.
4
Experimental methods give different types of data

• ChIP-chip
• microarray data defining genomic regions
• probe (with position usually defined)
  expression
• ChIP-Seq
• high throughput DNA sequence
• ACGATGTCA sequence fragments (from
  Solexa/SOLID/454)?
• sequence position undefined (search required)?
• The same issues exist for microarray vs. deep
  sequencing in gene expression experiments
• coverage
• cost
• practicality?

5
Raw (sequence) data

• Flat files, processed from base-calls to fasta
  format
• Solexa
• 25-30 bp reads
• Barcode is used to pool samples in one sequence
  run
• ACGT Expt1
• TGAC Expt2
• ACGTSequence
• TGACSequence

6
Choice of experiment

• Choice of experiment depends on the focus you
  require
• Whole genome broad coverage (of known genome)?
• or focused genomic region?
• or discovery based (known or unknown genome)?
• How much coverage do you need?
• Fewer broad experiments vs. many focused
  experiments?
• Custom chips can be easily designed for focused
  regions and custom applications.

7
Chip- Workflow
CHip-chip
CHip-seq

• Select antibody
• Decide how deep to sequence ( vs. coverage)?
• Sequence fragments
• Map Sequence to genomic position (BLAST/BLAT)?
• Identify peaks from data and minimise false
  positives
• Analyse peaks to predict binding sites

• Select antibody
• Select chip or design and select probes
• Map Array probes to genomic positon (BLAST/BLAT
  or lookup table from chip supplier)?
• Identify peaks from data and minimise false
  positives
• Analyse peaks to predict binding sites

8
Chip- Ringo Workflow example
9
Chip- output

• Peaks on the genome
• Score for each genomic position

BMC Bioinformatics 2007, 8219
10
Antibody selection

• Success depends on your antibody
• Select antibodies that are suitable for CHip-chip
  experiments
• Only a small number so far!
• List available from
• http//www.chiponchip.org/antibody.html

11
Microarray companies

• DNA microarrays suitable for ChIP-chip assays
• Affymetrix
• Human Chr2122 tiling microarrays
  (oligonucleotide arrays)?
• Human ENCODE tiling arrays (oligonucleotide
  arrays)
• Agilent
• Custom oligonucleotide arrays
• Nimblegen Systems, Inc.
• Human promoter microarrays
• Human ENCODE microarrays
• Custom oligonucleotide arrays
• Aviva Systems Biology
• Hu5K promoter arrays (PCR product arrays)?
• Hu20K promoter arrays (Oligo arrays)

12
Probe Design

• Tiling
• high-resolution arrays
• target genomic regions of interest
• whole genome or specific targeted regions?
• Agilent eArray probe database
• gt21 million tiled CGH and ChIP-on-chip probes
• Do it yourself
• unassembled genomes, etc...?

13
Mapping to genome

• The genome is still not constant, especially for
  many organisms
• You must map the probe/sequence to genomic
  location using
• standard alignment software (BLAST/BLAT/vmatch/...
  )?
• or rely on datafiles from the vendor (reccomended
  for most cases)?
• R packages exist for annotating probes to genomic
  location

14
Mapping to genome

• For sequence based methods this step is critical
  (and slow)?
• need unix server to run (or VMware)
• Do I need access to a computing cluster?
• choice of parameters for short sequences
• Filter raw sequences -gt representative sequence
  set
• Do I need to pre-filter data (some seqs will
  account for most of the compute time)?
• must be aware of speed vs. specificity for large
  datasets

Genome
15
Normalisation

• A normalization procedure
• (a) The MA plot before normalization shows a need
  for rotation to correct dye-bias.
• (b) To determine the correct angle of rotation,
  the s(M) vs s(A) plot of the differences between
  probes is generated This circumvents the effect
  of binding signal in determining the rotating
  angle for original MA plot in (a).
• (c) The MA plot after rotation by the angle
  determined in (b). The green line is the fitting
  line after rotation.
• (d) The MA plot after normalization..

• BMC Bioinformatics. 2007 8 219.

MA plot is a scatterplot with transformed axes.
The X-axis represents the average log intensity
from 2 channels while Y-axis represents the
log-ratios.
16
Peak detection

• What regions of DNA contain signal peaks?
• How to define a statistically significant peak?

Zheng, M. et al. (2007) ChIP-chip data, model,
and analysis. Biometrics, Vol 63, 787-796.
17
Normalisation

• Before normalization
• the mock control appears to show the same
  differential enrichment between genic and
  intergenic regions as the histone occupancy,
  suggesting that the differential enrichment may
  be an artifact.
• After normalization
• the mock control no longer shows significant
  differential enrichment while H3 and H4 profiles
  still do

• Peng et al. BMC Bioinformatics 2007 8219
  doi10.1186/1471-2105-8-219

18
Noise

• Contamination
• Do sequences match the expected genome?
• Sequencing errors
• Can you determine where a sequencing error is?
• Multiple-mapping sequences
• Many sequences do not unique genome matches
• Dye specific bias
• ChIP-chip data for chromatin-associated proteins
  and histone modifications present additional
  challenges
• as they often display broad regions of
  enrichment. This is in contrast to the isolated
  and sharp peaks that are typical for the binding
  of transcription factors.

19
Peak detection - replicates

• Use replicates to improve detection
• Peaks that are consistent between replications
  are more likely to be true

Zheng, M. et al. (2007) ChIP-chip data, model,
and analysis. Biometrics, Vol 63, 787-796.
20
What next?

• Given that you've identified accessible regions
  in the genome
• What information can be gathered from this
  sequence?
• Use discovery methods to look for common patterns
  in the regions
• MEME, etc
• Use TFBS databases to look for known
  transcription factor binding sites in the
  sequence
• Transfac
• High coverage
• Noisy database
• Jaspar
• Low coverage
• Higher quality?

21
R packages for chip-chip

• Ringo
• Well documented workflow and good tutorial
• BAC
• Perfect example of minimal documentation
• Bayesian Analysis of ChIP-chip data

22
Summary

• You should be able to
• Understand the differences between CHip-chip and
  CHip-Seq and identify key decision making steps
  for choosing a platform
• Identify bioinformatics requirements for handling
  CHip-chip and Chip-Seq datasets
• Find transcription factor binding sites in
  genomic data
• Understand the need for skills in handling large
  datasets