| |
 | |
 | |
 | |
 | |
 | |
 | Login |
 | |
| Overview | |
| |
| Quality Assurance Methods for Processing Microarray Imagery | |
| |
| Using Visualization for Microarray Data Analysis | |
| |
| Visualization of Soybean Production | |
| |
Contacts: Peter Bajcsy; pbajcsy@ncsa.uiuc.edu
Collaborator: Lei Liu, The W. M. Keck Center for Comparative and Functional Genomics
Collaborator: Lei Liu, The W. M. Keck Center for Comparative and Functional Genomics
|
Problem Definition This project began with a body of DNA microarray data that needed to be analyzed. Because of the complexity of the data, data mining and visualization of microarray features is essential to any interpretation of a microarray experiment. The objective for this project was to apply proven data mining techniques in the domain of genetics, develop novel visualization/analysis methods and present the results to researchers in the field.  Approach For this project we addressed the issue of fast screening and inspection of DNA microarray data by using novel image-based visualization approaches, including visualization of input image channels, grid alignment results, screening results, high dimensional features, and classification labels. The novelty came in viewing the DNA microarray data as high-dimensional images throughout the entire analysis process. For example, laser scanned imagery, extracted features and labeled classification results were represented visually rather than numerically. This was possible because microarray dots form a grid pattern, creating intrinsically grid-based information. Thus, it was natural to extract features while maintaining the grid. Each extracted feature formed a point in the feature image that was then used for classification and visual inspection. From a data analysis perspective, this type of display was very suitable for screening errors and inspecting analyzed data. Although extracted features from DNA microarray scanned imagery might not be spatially related, we surmised that it could be beneficial to introduce a spatial pattern of expected up-regulated and down-regulated genes into the design arrangement on a microarray glass slide. We hoped this would provide a good method for performing fast image calibration. Results were positive, demonstrating the benefits of spatially related microarray grid information. One benefit of image-based visualization is the intuitive display of features. Extracted features are usually presented in table form with multiple variables associated with each grid location; for instance, mean, median, standard deviation, and ratios. A high-dimensional image (x, y, feature value) provides a 3D data cube with each image band (frame or 2D slice) able to illustrate variations of features over the entire set of extracted features. For example, a fast detection of systematic errors can be conducted by visual inspection of several feature bands. In addition to visually exploring extracted features, image-based visualization provides a means for efficient inspection of labeled classification results. For instance, a result of K-means clustering can be easily displayed with color labels as one resulting image. The results of hierarchical clustering methods, e.g., single-link or complete-link clustering, can be shown as a cross section of a labeled 3D cube (x, y, clustering level) instead of a standard dendogram-based visualization. Finally, this project demonstrated the use of image-based visualization for exploration of experimental variables. In order to investigate a functional dependency of a gene expression level on a selected variable, one can stack classification results into a 3D data cube (x, y, variable value) and view a 2D cross section of the multi-grid data cube along the variable axis. This type of visualization serves as a good inspection tool while conducting gene correlation studies.  |
|
 Results This project was an unfunded research project with the W. M. Keck Center for Comparative and Functional Genomics. Proposals have been submitted to continue this promising research. Publications Peter Bajcsy and Lei Liu, "An Image-Based Visualization of Microarray Features and Classification Results," Poster Proceedings of the 10th Intelligent Systems for Molecular Biology, Edmonton, Alberta, Canada, 3-7 August 2002, pp.59 Download Document (PDF 3.9 MB) |
