Microarray Image Processing

1
Microarray Image Processing
Karl Fraser, Zidong Wang, Yongmin Li, Paul Kellam
and Xiaohui Liu
School of Information Systems, Computing and
MathematicsBrunel University, London, UK
2
Presentation Outline

• Microarray Overview
• The Problems
• Background Reconstruction
• Proposed Process
• Results
• Future Work
• Questions

3
Microarrays

• The Highlights

4
Microarray Background

• Individual genes can be compared using a
  Competitive Hybridisation
• Microarrays allow this experiment to be carried
  out on a mass scale at a microscopic level
• Typically 6-30 thousand genes can be analysed on
  one chip simultaneously

5
Ratio Calculation

• For each gene, 4 values calculated
• Cy5 Signal, Cy5 Noise (Red Control)
• Cy3 Signal, Cy3 Noise (Green Test)
• Expression Ratio
• Shows how much the gene was expressed in infected
  and normal strains

6
The Problem

• What are the problem characteristics

7
The Problem

• GenePix Method

8
The ProblemPixel Relations
9
The ProblemChannel Biases
10
The ProblemData Characteristics

• Printing

• Cleaning

• Background

• Shapes

11
Data Characteristics...cont

• Large variation in noise and source
• Interesting genes typically have very low
  signal-to-noise ratios
• Large amounts of missing data are common
• Experiments often contain few, if any technical
  repeats

12
Overview of Proposed Process

• Fourier Inpainting

13
Reconstruction

• Mask out gene spots
• Estimate pixels behind gene, based on surrounding
  region
• Onion, Nearest Neighbour, etc
• Could choose to blur estimation to
• reduce individual pixel error
• Subtract estimation from original image and
  calculate gene ratio

14
Contemporary Techniques

• Bertalmio
• Chen
• Oliveira

15
Contemporary Techniques...cont

• ONeill et al. 2004
• Progressively build estimation of noise behind
  gene spot
• Uses technique taken from texture estimation
  (Efros 1999)
• Bias estimation towards the median background to
  reduce the effect of texture estimation anomalies
• Algorithms takes 2hr per slide

16
Fourier Inpainting
17
Results
18
Results...cont
19
Results...cont
RAW
FFT
ONeill
20
Results...cont
21
Results...cont

• Gene banding at 38 and 1519 associated with
  saturated controls
• Partial experiment banding at 7, 9 and 25
  associated with poor surfaces

22
Conclusions

• Background sample details gene intensity
  variation
• Low signal-to-noise genes processed more
  accurately
• Fast microarray reconstruction
• Results highlight techniques have good potential
• Not perfect reconstruction, areas of improvement

23
Future Work

• Optimisation (MATLAB, 30 decrease)
• Test biological significance
• More complex methods of TE such as MRF or
  Wavelets
• Affymetrix Protein arrays
• A Hybrid FFT / Histogram based reconstruction
  system could be benificial

24
References

• Image Inpainting
• M. Bertalmio et al., Proc. Graphics Interactive
  Techniques , 2000
• Fast Digital Image Inpainting
• M. M. Oliveira et al., Proc. Visualization, Image
  Processing, 2001
• Improved Processing of Microarray Data Using
  Image Reconstruction Techniques
• ONeill et al., IEEE NanoBioscience, Dec 2003

25
Acknowledgments

• Data kindly provided by
• Dr. Kellam from the Dept. of Immunology and
  Molecular Pathology, University College, London
• Dr. Li from the Dept. of Biological Sciences,
  Brunel University, Uxbridge

26
Questions
27
(No Transcript)