Close-to-the-experiment Data analysis

1
Close-to-the-experiment Data analysis
Dr. Werner Van Belle werner.van.belle_at_gmail.com,
werner_at_onlinux.be
2
Bridging The Gap

• Core of most biological research is experimental

Interpretation
Data Reduction
Experiments
3
Bridging The Gap

• Core of most biological research is experimental

Interpretation
Data Analysis
Experiments
4
Contents

• Part 1. 2DE Gel Correlation Analysis
• Part 2. Maldi Tof Artefacts / Denoising
• Part 3. Accuracy Analysis of a Micro-Array
  Experiment
• Part 4. Protein Interaction Map Integration

5
Part 1. 2DE Gel Analysis
Werner Van Belle werner.van.belle_at_gmail.com,
werner_at_onlinux.be In cooperation with Bjørn Tore
Gjertsen, Nina Ånensen Ingvild Haaland, Gry
Sjøholt, Kjell-Arild Høgda
6
2D Gels
Patient 2Age 46
Patient 1Age 57
Courtesy Gry Sjøholt, Nina Ånensen Bjørn Tore
Gjertsen
7
Initial Problem

• The question we were asked
• Is there a relation between various parameters of
  AML/ALL cancer patients and their P53
  biosignatures / isoforms ?
• Gels /- 97 gel images of different patients
• Biological Parameters
• FAB Classification (AML/ALL), AML Class, Flt3
  (WT/ITD)
• Resistance AML, Resistance ALL, Survival AML,
  Survival ALL
• BCL2, Stat5 GMCSF, Stat3 IL3, Stat1 Ifng, CD4, C34

8
Standard Solution

• Detect Spots, Measure Spot Volumes, Compare
• Non Trivial Solution
• Spot identity unknown, often no calibration spots
• Manual interpretation dangerous shifts of spots
  are difficult to interpret
• Some PTM influence spot positioning, complicating
  the matter

Complicated method Tedious work Lousy results
9
Manual Comparison
10
2D Gel Analysis

• Step 1 Image registration rotate, scale
  translate

11
Different Operations

• Scaling (Zoom)
• Rotation
• Translation

Calibration spots Antibody spots Manual annotation
Image registration techniques Geocoding Landmark
tracking Standard spot detection
12
Pairwise Image Alignment
13
Noise hinders pairwise alignment
14
1. Artefacts in 2D gels
Camera Noise
Gels have been altered to respect NDA
15
2. Artefacts in 2D Gels
Camera Warping
16
3. Artefacts in 2D gels
Inconsistencies over different machines
KODAK Image Station
Typhoon Image Station
Courtesy Gry Sjøholt, Nina Ånensen Bjørn Tore
Gjertsen
17
4. Artefacts in 2D gels
Underexpressed tails
Gels have been altered to respect NDA
18
5. Artefacts in 2D gels
Non linearity
Gels have been altered to respect NDA
19
6. Artefacts in 2D gels
Washing
Gels have been altered to respect NDA
20
7. Artefacts in 2D gels
Drying
Gels have been altered to respect NDA
21
8. Artefacts in 2D gels
Dots
Gels have been altered to respect NDA
22
9. Artefacts in 2D gels
Poor XY resolution
Poor grey-value resolution
Gels have been altered to respect NDA
23
10. Artefacts in 2D gels
Clipping
Gels have been altered to respect NDA
24
11. Artefacts in 2D gels
Unclean Lenses
25
Denoising 2DE gels
Before
After
26
Denoising I
Input Image
27
Denoising II
Inverted Image
28
Denoising III

• Calculate Background Variations

29
Denoising IV

• Remove backround variation

Original
Divide original by background variation image
Denoised
30
Denoising V

• Thresholded

Clip everything gt 1
31
Denoised result

• Salt Pepper removal

Median Filtering
32
Original
33
Denoising enables pairwise alignment
34
Outline

• Correlation analysis
• Requires multiple aligned gels
• Multiple gel alignment based on pairwise
  alignment
• Pairwise alignment difficult due to many
  artefacts
• Developed denoising algorithm
• Pairwise alignment possible
• How to align multiple gels ?

35
Cummulative Superposition

• Idea
• take first gel, superimpose second gel
• take third gel, superimpose on projection of
  previous gels
• repeat process for all gels

This does not work, we merely find a suitable
superposition to reflect the first images.
36
Cummulative Superposition
Final Overlay Image
Initial 2DE Gel Image
37
Cummulative Superposition
Final Overlay Image
Initial 2DE Gel Image
38
Multi Gel Alignment

• 1- align all image pairs -gt X.X alignments
• 2- find an optimal (x,y) position that minimizes
  the overall alignment error

100 images at 1024 x 1024 65011712 operations per
cross correlation 5000 cross correlations 32505856
0000 operations in total 325.109 FLOP
theoretical 2.7 hours practical 3 days
39
2D Gel Overlays

• Superposition of all images

Mother image
40
2D Gel Overlays
Reflects Known Protein Isoforms
41
Step 1 Alignment
42
2D Gel Analysis

• Step 2 Intensity Normalization

43
Background Differences
44
Background Differences
45
Step 2a Background Intensity
46
Contrast
47
Contrast
48
Step 2b Intensity Normalization
49
2DE Gel Analysis

• Step 3 Correlate

50
Step 3 Correlation
51
Step 3 Correlation
52
Initial Problem

• Is there a relation between various parameters of
  AML/ALL cancer patients and their P53 isoforms ?

53
P53 Biosignatures vs Age
54
2D Gel Analysis

• Step 4 Mask

55
Step 4 Masking
56
Step 4a Significance
57
Significance Mask
58
Step 4b Variance
59
Variance Mask
60
Step 4c Overall Mask
61
Overall Mask
62
P53 Biosignature vs Age
63
Step 5 3D Visualization
64
Step 5 3D Visualization
65
Resource Usage

• 132 Parameters, 13 correlation sets, 128 images
• Creating the fine-tuned overlay alignment 72h
• Computing all the correlations 85.55h, which
  produced 5.8 Gb of raw data.
• Rendering of the movies 5 hours per movie, with
  1416 images 7080h

66
Part 2. Maldi-TOF Artefacts
Werner Van Belle werner.van.belle_at_gmail.com,
werner_at_onlinux.be In cooperation withOlav
Mjaavatten, Kari Espolin Fladmark Stijn Ove
Døskeland
67
MALDI
Laser
Protein
Digestion (Trypsine)
Sublimation
Crystalisation
Matrix absorpbing wavelenght x
Charged particles
Time of flight
Protein fingerprinting
Detection
Mass spectrum
Peaklist
Sequencing
Diagonalisation
68
MALDI
69
Artefacts I

• Static Noise

Mass spectrum output
Frequency Analysis
70
Artefacts II

• Sweeping Tones in LIFT

Mass spectrum output
Frequency Analysis
71
Artefacts III

• Coherent Pulses

1 Shot
10 Shots
Mass spectrum output
Frequency Analysis
72
Artefacts III

• Coherent Pulses

100 Shot
1000 Shots
Mass spectrum output
Frequency Analysis
73
Denoising method

• Multi-rate spectral analysis excelent tool for
  event detection

74
Denoising method
Initialisation
Haar wavelet Decomposition
75
Denoising method
Normalisation Event Enhancement
Wavelet Composition Saving
76
Wavelet Enhancement Global
Original
Denoised
Frequency Analysis
Mass Spectrum
77
Wavelet Enhancement Local
Original
Denoised
Frequency Analysis
Mass Spectrum
78
Part 3. Micro-Array Accuracy Analysis
Werner Van Belle werner.van.belle _at_ gmail.com,
werner _at_ onlinux.be In cooperation with Nancy
Gerits, Ugo Moens, Halvor Grønaas, Lotte Olsen,
Ruth Paulssen
79
Intensity Distribution
80
Intensity Distribution
Gating ? Critical Mass ?
81
Cy5 ? Cy3
82
Measurement Accuracy
Green
Relative Error
PD of error at distance z
PD of error at distance y
Red
PD of error at distance x
83
Intensity Dependent Error Distribution
84
Intensity Dependent Error Distribution
In 95 of the cases the measurement error will
fall within -49364936
85
Confidence Interval for 1 Spot
In 95 of the cases, the actual value will Range
within the measurement -49364936
86
Multiple Spots

• Multiple measurements lead to better estimates /
  smaller confidence intervals

87
Reported Regulations
88
Omitted spots too close to error
89
Part 4. Protein Interaction Map Integration
Werner Van Belle werner.van.belle_at_gmail.com,
werner_at_onlinux.be In cooperation with Nancy
Gerits, Ugo Moens
90
Gene Expression
91
Influenced by/Influences

• MK5 -gt Multiple changes in gene expression
• 27000 gene expressions measured
• Those that change will very likely influence
  other proteins

Which proteins are likely influenced by our
measured up/down regulations ?
92
The 'Involved' Game

• Protein change will influence nearby proteins,
  which in turn ...

1.0
0.8
0.6
0.6
93
The 'Involved' Game

• Multiple proteins changes will all influence
  their neighbors as well.

1.0
1.6
1.4
1.0
94
The 'Involved' Game

• This network is iterated a number of times to
  expand the sphere of influence of all the altered
  gene expressions.
• affected proteins will have higher numbers
• Protein Interaction key mechanism for signal
  transduction
• Protein Interaction Network as published by

Jean François Rual et al. Towards a Proteome
Scale Map of the Human Protein Protein
Interaction Network Nature 2005 vol 437, p.
1173-1178
95
Involved Proteins by Rank
96
Involved Proteins by Rank
97
Involved Proteins by Rank
98
Involved Proteins Network
99
Involved Proteins Network

• Red Highest involvement Blue Lowest
  Involvement
• Based on our lowest estimates for up/down
  regulation
• Based on the high confidence set of protein
  interactions
• Measured gene expressions are not listed

100
Involved Proteins Network
101
Involved Protein Network
102
Involved Protein Network
103
Involved Protein Network
104
Credits

• 2DE Gel Imaging and Patient sampling
• Nina Ånensen, Bjørn Tore Gjertsen, Ingvild
  Haaland, Øystein Bruserud, Gry Sjøholt
• Maldi TOF Mass Spectra
• Olav Mjaavatten, Kari Espolin Fladmark, Stijn Ove
  Døskeland
• Micro-arrays
• Nancy Gerits, Ugo Moens, Halvor Grønaas, Lotte
  Olsen, Ruth Paulssen