Rahnenfueher, Joerg - Hybrid clustering for microarray image analysis combining intensity and shape features

ECCB 2002 Poster sorted by: Author | Number

Next | Previous poster (in order of the view you have selected)

Title: Hybrid clustering for microarray image analysis combining intensity and shape features	P126
Rahnenfueher, Joerg rahnenfj@stat.berkeley.edu Department of Statistics, University of California, Berkeley; Center for Human Molecular Genetics, University of Nebraska Medical Center

Image analysis is the first crucial step to obtain reliable results from cDNA microarray experiments. It can be divided into three basic steps. First, target areas belonging to single spots have to be identified. Then, those target areas are partitioned into foreground and background. Finally, two scalar values for the intensities are extracted. These goals have been tackled either by intensity histogram methods (see Chen et al., 1997 for an early reference) or by spot shape methods (for example Yang et al., 2001). However, it would be desirable to have hybrid algorithms that combine the advantages of both approaches.

A new robust and adaptive histogram type method is pixel clustering, which was applied for detecting and quantifying microarray spots (Bozinov and Rahnenführer, 2002). It uses a robust version of the clustering algorithm PAM (Kaufman and Rousseeuw, 1990) to distinguish between foreground and background pixels. In addition, a more time efficient version based on k-means was developed. On real microarray spots both algorithms produce reliable results for spots with typical impairments. It was also shown that the physical size of the target area barely influences the final intensity estimates.

Here, this histogram method is extended to a hybrid algorithm, where the spot shape is effectively integrated. A bivalence mask is constructed by overlaying the binary clustering results for all single spots. It estimates the expected spot shape and is used to filter the data, improving the results of the cluster algorithm. Using the indicator values of the clustering results instead of original intensity values guarantees robustness of this step. The introduction of a third class of pixels that are assigned neither to foreground nor to background makes the algorithm more flexible and allows a better treatment of low quality areas on the microarray slide. The 'stability' is introduced as a suitable quality measure. It is defined as the relative frequency of pixels in the foreground area that are not deleted by mask matching.

To demonstrate the practical feasibility of this proceeding, the full algorithm is evaluated on a real data set. Figure 1 shows a microarray spot with a large artifact on top. Figure 2 shows the pixel selection for this spot, based on the hybrid algorithm. Black pixels represent the local background and white pixels the spot foreground. Pixels that belong to the artifact are eliminated in an appropriate manner. The proposed algorithm represents a true hybrid microarray image analysis solution that incorporates both shape and histogram features. It is specifically adapted to deal with typical microarray image characteristics.

(For the figues see the web representation of the poster abstract)

[1] Chen, Y., Dougherty, E.R. and Bittner, M.L. (1997) Ratio-based decisions and the quantitive analysis of cdna microarrays, Journal of Biomedical Optics, 2, 364-374.
[2] Bozinov, D. and Rahnenführer, J. (2002) Unsupervised technique for robust target separation and analysis of microarray spots through adaptive pixel clustering, Bioinformatics, 18, 747-756.
[3] Kaufman L. and Rousseeuw, P.J. (1990) Finding groups in data: An introduction to cluster analysis, Wiley, New York.
[4] Yang, Y.H., Buckley, M.J. and Speed, T.P. (2001) Analysis of cDNA microarray images. Briefings in Bioinformatics, 2(4), 341-49.