mixture model analysis of dna microarray...
TRANSCRIPT
![Page 1: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/1.jpg)
Mixture Model Analysis of DNAMicroarray Images
Blekas, Galatsanos, Likas and Lagaris
Presentation by Warren Cheung
1
![Page 2: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/2.jpg)
Overview
• Data — DNA Microarrays
• Segmentation Tasks
– gridding
– spot analysis
• Experimental Evaluation
2
![Page 3: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/3.jpg)
DNA Microarrays
• DNA “chip” experiments
• samples of DNA (RNA, protein, etc...) placed on the chip
• “wash” with reporter molecules
– fluorescent in two colors (red and green)
– each reporter binds with a unique target
• take image
3
![Page 4: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/4.jpg)
Result
• intensity of color indicates amount of target present
• “high-thoroughput” — can do many spots on a tiny chip
• need software to analyse these images
4
![Page 5: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/5.jpg)
5
![Page 6: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/6.jpg)
Two Segmentation Problems
• gridding: separate the spots from each other (global and localsegmentation)
• spot analysis: separate the spots from the background(Gaussian Mixture Model)
6
![Page 7: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/7.jpg)
Gridding
• generates a rectangular, axis-aligned boundary for each spot
• First pass: Global boundary
– for each row, sum the intensities of every pixel in the row
– find the minimum value between peaks, these are boundaries
– repeat for the columns
7
![Page 8: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/8.jpg)
Gridding (2)
• Second pass: refine the boundaries
– consider two side-by-side spots
– within the global row boundaries for these two spots, findsum of pixel intensities for each column
– move the column boundary to the minimum
– repeat for rows
8
![Page 9: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/9.jpg)
9
![Page 10: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/10.jpg)
Spot Analysis
• wish to label each pixel
• (B) Background
• (F) Foreground
• (A) Artifact
10
![Page 11: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/11.jpg)
Gaussian Mixture Model
• treat this as a clustering problem
• use a mixture of K density functions
• K = 2: (B) and (F)
• K = 3: (B) and (F) and (A)
• use EM and MAP to optimise
• use cross-validated likelihood to choose K
11
![Page 12: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/12.jpg)
EM: Expectation Maximisation
• maximization of the likelihood
• start from an initial guess
• iteratively improve the model
• E-step: compute “likelihood”
• M-step: find “best” modification of model parameters
12
![Page 13: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/13.jpg)
EM on the basic model
• Basic model: probability that pixels with a certain intensityhas a particular label is given by a Gaussian
• E-step: given current model of Gaussians (µ(t)j ,Σ(t)
j ), computeposterior
• M-step: update Gaussians (compute new mean, covariancematrices) and mixing weights
13
![Page 14: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/14.jpg)
Cross-Validated Likelihood
• partition the data into u sets Xs = X1, . . . , Xu
• for each Xs compute the model using the data X −Xs
• check by computing the likelihood of Xs
• get an average score over the u sets
• repeat for K = 3
14
![Page 15: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/15.jpg)
MAP (Maximum a posteriori) estimation
• wish to take advantage of spatial information
• add an additional probability distribution — probability that aposition has a particular label
• use a Markov random field model
• maximise the (log) density function using EM
• M-step involves solving quadratic programming (QP) problem
15
![Page 16: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/16.jpg)
Evaluation of Experiments
• artificially created images
• publicly available microarray databases
16
![Page 17: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/17.jpg)
Gridding
• evaluated by visual inspection (“gold standard”)
• 3 grades: Perfect (entire spot), marginal (> 80%) or incorrect
• compare against 2 other tools
Proposed Spotfinder Scanalyze
Perfect 89.6 72.8 48.7
Marginal (> 80%) 9.2 14.3 22.6
Incorrect 1.2 12.9 28.7
17
![Page 18: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/18.jpg)
Spot Analysis - Artificial Data
• compare as noise is increased
• evaluate percentage of misclassified pixels and error incomputed spot intensity
• as true result is known, they can calculate error
18
![Page 19: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/19.jpg)
19
![Page 20: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/20.jpg)
Spot Analysis - Real Data
• compare spot intensity to other methods and existing tools
• argue that values are more representative of spots
20
![Page 21: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/21.jpg)
21
![Page 22: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/22.jpg)
22
![Page 23: Mixture Model Analysis of DNA Microarray Imagesdnahelix.wdfiles.com/local--files/cmpt-880/cs880-gaussian-slides.pdf · Mixture Model Analysis of DNA Microarray Images Blekas, Galatsanos,](https://reader033.vdocuments.us/reader033/viewer/2022060302/5f08a6567e708231d4230d4c/html5/thumbnails/23.jpg)
Conclusion
• gridding: Exploit feature of experimental layout and knowledgeof results to do rapid segmentation
• spot analysis: Iteratively improve a mixture model to fit thedata
23