2005 Fiscal Year Final Research Report Summary
Network Analysis from microarray data
| Project/Area Number |
12206008
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research on Priority Areas
|
| Allocation Type | Single-year Grants |
|---|
| Review Section |
Biological Sciences
|
|---|
| Research Institution | Kyushu University |
|---|
Principal Investigator |
KUHARA Satoru Kyushu University, Faculty of Agriculture, Professor, 大学院農学研究院, 教授 (00153320)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
TASHTO Kousuke Kyushu University, Faculty of Agriculture, Associate Professor, 大学院農学研究院, 助教授 (00192170)
ONISHI Sadanori Kyushu University, Faculty of Agriculture, Professor, 大学院数理学府, 教授 (40090550)
MARUYAMA Osamu Kyushu University, Faculty of Agriculture, Associate Professor, 大学院数理学府, 助教授 (20282519)
|
|---|
| Project Period (FY) |
2005
|
| Keywords | Bioinformatics / Genome / Informatics / Expression control / Microarray |
|---|
| Research Abstract |
Microarray techniques provide new insights into molecular classification of cancer types, which is critical for cancer treatments and diagnosis. Recently, an increasing number of supervised machine learning methods have been applied to cancer classification problems using gene expression data. Support vector machines (SVMs), in particular, have become one of the most effective and leading methods. However, there exist few studies on the application of other kernel methods in the literature. We apply a kernel subspace (KS) method to multiclass cancer classification problems, and assess its validity by comparing it with multiclass SVMs. Our comparative study using seven multiclass cancer datasets demonstrates that the KS method has high performance that is comparable to multiclass SVMs. Furthermore, we propose an effective criterion for kernel parameter selection, which is shown to be useful for the computation of the KS method.
Microarrays pose a great challenge on the data analysis, bec
… More
ause the number of genes often exceeds tens of thousands, whereas the number of samples available is at most a few hundred. In microarray data analysis, gene selection has been a central issue in recent years. Gene selection plays several important tasks. If it is possible to identify a small subset of biological relevant genes, it may provide insights into understanding the underlying mechanism of a certain biological phenomenon. In general, gene selection can be performed in different manners: the filter and wrapper approaches. We propose a new filter approach to gene subset selection for kernel-based methods. Our major goal is to enhance the performance of kernel-based classifiers without resorting to the wrapper approach. This can be realized by direct performing classification in the same feature space. It is different from the conventional filter approach combined with kernel-based classifier, where gene subset selection is performed in input space. We show that several well-known class separability criteria can be kernelized. Gene subset selection is performed based on the kernelized criteria. Our strategy is assessed by combining it with three kernel-based classifiers, from simple to advanced ones. We apply it to cancer classification problems on acute lymphoblastic leukemia and demonstrate validity of our proposed strategy by comparing it with several methods. Less
|
