Predicted hydroxyl radical cleavage data for ENCODE regions

1
Predicted hydroxyl radical cleavage data for
ENCODE regions

• Tullius Lab
• Jay Greenbaum, Steve Parker, Tom Tullius
• Boston University
• Chromatin Chromosomes conference call
• July 13, 2005

2
Hydroxyl radical cleavage of DNA
3
Prediction algorithm
4
Prediction accuracy
5
ENr221
6
Data segmentation

• Windowing method
• Demarcate regions of high low intensity
• Look for enrichment vs. other datasets
• HMM (Bill Noble)
• 2-5 state models
• 1nt 50nt resolution

7
Enrichment analysis (Yutao Fu Zhiping Wengs
Lab)

• Determine overlap of nucleotides between datasets
• Determine overlap of nucleotides between shuffled
  datasets
• Calculate ratio

8
Enrichment analysis cont.
9
HMM Segmentations

• n-state totally connected HMMs
• Single nucleotide resolution (n2,3)
• 10bp non-overlapping windows (n2-5)
• 50bp non-overlapping windows (n2-4)
• Applications
• Alignment of OH cleavage intensity (state
  sequences)
• Promoter sequences, etc.
• Cross-species comparisons
• Clustering of cleavage patterns
• Overlapping windows of state sequences
• Cluster to find related regions

10
Distribution of cleavage intensities
11
Single nucleotide 3-state HMM state composition
12
State approximation
13
HMM Segmentations

• ENr221 revisited

14
Observations

• Regions of high cleavage intensity enriched
  around BU-TSS, GC-TSS, DNase HSS, CpG
• HMM a good starting point, but must be refined to
  better model the data

15
Future work

• Mine segmented data
• Determine state enrichment vs. other data
• Attempt clustering of overlapping n-mers
• More specific HMMs
• Promoter regions, nucleosome positioning
  sequences, etc.
• Data segmentation for other species comparison
  to human
• Perform cleavage experiment on ENCODE region(s)
  with Beckman CEQ 8000 capillary sequencer

16
Acknowledgements

• Bill Noble (Univ. Washington)
• Yutao Fu (BU)