Identifying Patterns in DNA Change

1
Identifying Patterns in DNA Change
Jason Gilder Bioinformatics Research Group Wright
State University MAICS Presentation April 12,
2003
2
ALU Background

• Well-known sequences broken into families
• Short Interspersed Repetitive Elements (SINEs)
• Approximately 280 bp long
• 10 of human genome

3
ALU Background Cont

• Proliferate through retrotransposition
• - Copy is transcribed, reverse transcribed,
  and
• reinserted at distant site
• Original progenitor sequence known
• Can trace evolutionary path
• - Number of changes
• - Types of changes

4
Problem

• Use EC to predict substitution rates
• (e.g. The number of Cs that used to be As)
• Only use features of repeat itself, not
  progenitor
• - Content information for repeat
• - GC content in flanking regions
• Feasible?
• - Enough features?
• - Correct features?

5
Feature Set

• 16 Features
• - Length of repeat, of As, of Gs, of
  Cs,
• of Ts, and GC Content Percentage of ALU
• GC content for 10 flanking regions
• 
  (500 20,000 nts)

6
Data Set

• ALU Y Family in Chromosome 1
• CENSOR used to get substitution data
• http//www.girinst.org/Censor_Server.html
• 6,749 examples
• 5,000 Training ( 74 )
• 1,749 Holdout Testing ( 26 )
• Each chosen randomly

7
CENSOR Alignments
ALUY 11 282 CONTIG-1P1 317456 317723
0.88 0.10 1.53 272 206.52
GGTGGCTCACGCCTGTAATCCCAGCACTTTGGGAGGCCGAGGCGGGCGGA
TCACGAGGTCAGGAGATCGA

GCTGGATGTC-CCTGTAATCCCAGCACTTTGGGAGG
CCGAGGCGGGTGGATCATGAGGTCAGGAGATCGA
GACCATCCTGGCTAACACGGTGAAACCCCGTCTCTACTAAAAATACAAAA
AATTAGCCGGGCGTGGTGGC

GACCATTCTGGCTAACACAGTGAAACCCCGTCTCTACTAAAAATACAAAA
AATTAGCCAGGCATGGTGGC GGGCGCCTGTAGTCCCAGCTACT
CGGGAGGCTGAGGCAGGAGAATGGCGTGAACCCGGGAGGCGGAGCTT

ACACGCCTCTAGTCCCAACTA
CTCAGGAGGCTGACACAGGAGAATCACTTGGACCCGGGAGGTGGAGGTT
GCAGTGAGCCGAGATCGCGCCACTGCACTCCAGCCTGGGCGACA
GAGCGAGACTCCGTCTCA

GCAGTGAGCTGAGATCACGCCACTGCACTCCAGCCTGGGTGA-AAA--GA
GACTCCGCCTCA Containing 239 matches, 3 gaps
and 29 mismatches including 19 transitions
8
Genetic Programming

• Equations built like parse trees
• Operator, input, and constant nodes
• (I6 35) 2.87

I6
35
2.87
9
GP Reproduction
Parent 1
Parent 2



I2
I4
2.87
I6
35
Child 2
Child 1



2.87
I4
I2
I6
35
10
Operator Node Set
Min returns the minimum of two nodes or
subtrees Max returns the maximum of two nodes
or subtrees   Cos if the connected nodes are x
and y, it returns x Cos(y)   Sin if the
connected nodes are x and y, it returns x
Sin(y)   Ave if the connected nodes are x and
y, it returns (x y)/2 Log if the connected
nodes are x and y, it returns x Log(y)
11
Mask Operator Nodes

• All features mutable binary mask
• Summation fi mi
•  
• Multiplication fi mi
•  
• SumSquareRoot fi mi

12
Initial GP Results

• Classifying C -gt G
• Fitness average absolute error
• Classification Rate 46
• Average absolute error 0.75

13
Theta Factor Offset

• Average absolute error 0.75
• If error lt 0.5, correct classification
• Subtract theta from solution to get correct
  solution
• Using 0.30, classification rates jumped to 66
• Linear search for best theta 0, 1 in 0.01
  increments

14
Initial Results
Progenitor Sequence
Alu
( training classification, test classification )
15
Regional GC Analysis

• All Experiments redone utilizing only GC flanking
  content.
• No ALU information used

16
Regional GC Analysis Results

• Features regional GC content

Progenitor Sequence
Alu
( Previous classification rates in parentheses )
17
Context Analysis Masking CpGs

• Substitution rates from Cs and Gs redone
• All CpGs were masked
• Removed some independent mutation factors

18
Masked CpG Results

• Features Flanking GC with masked CpGs

Progenitor Sequence
Alu
( Previous classification rates in parentheses )
19
Conclusions

• Successfully predicted substitution rates
• Regional GC Content holds needed information
• 10 / 12 rates gt 80
• 6 / 12 rates gt 90
• Future Work Classifying entire genome

20
Acknowledgements

• Dr. Dan Krane
• Dr. Travis Doom
• Dr. Michael Raymer

Dr. Mateen Rizki
Bioinformatics Research Group http//birg.cs.wrig
ht.edu
This work was supported in part by the National
Science Foundation (grant EIA-0122582), and by
the Dayton Area Graduate Studies Institute.