Identification of pathway and gene markers using enhanced directed random walk for multiclass cancer expression data
Cancer markers play a significant role in the diagnosis of the origin of cancers and in the detection of cancers from initial treatments. This is a challenging task owing to the heterogeneity nature of cancers. Identification of these markers could help in improving the survival rate of cancer patie...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/98108/1/NiesHuiWenPSC2020.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Cancer markers play a significant role in the diagnosis of the origin of cancers and in the detection of cancers from initial treatments. This is a challenging task owing to the heterogeneity nature of cancers. Identification of these markers could help in improving the survival rate of cancer patients, in which dedicated treatment can be provided according to the diagnosis or even prevention. Previous investigations show that the use of pathway topology information could help in the detection of cancer markers from gene expression. Such analysis reduces its complexity from thousands of genes to a few hundreds of pathways. However, most of the existing methods group different cancer subtypes into just disease samples, and consider all pathways contribute equally in the analysis process. Meanwhile, the interaction between multiple genes and the genes with missing edges has been ignored in several other methods, and hence could lead to the poor performance of the identification of cancer markers from gene expression. Thus, this research proposes enhanced directed random walk to identify pathway and gene markers for multiclass cancer gene expression data. Firstly, an improved pathway selection with analysis of variances (ANOVA) that enables the consideration of multiple cancer subtypes is performed, and subsequently the integration of k-mean clustering and average silhouette method in the directed random walk that considers the interaction of multiple genes is also conducted. The proposed methods are tested on benchmark gene expression datasets (breast, lung, and skin cancers) and biological pathways. The performance of the proposed methods is then measured and compared in terms of classification accuracy and area under the receiver operating characteristics curve (AUC). The results indicate that the proposed methods are able to identify a list of pathway and gene markers from the datasets with better classification accuracy and AUC. The proposed methods have improved the classification performance in the range of between 1% and 35% compared with existing methods. Cell cycle and p53 signaling pathway were found significantly associated with breast, lung, and skin cancers, while the cell cycle was highly enriched with squamous cell carcinoma and adenocarcinoma. |
|---|