A classification method of gene expression profile based on a locally linear embedding algorism with improved distance.
- Author:
Xianfa CAI
1
;
Jia WEI
;
Guihua WEN
;
Jie LI
Author Information
1. Medical Information Engineering School, Guangdong Pharmaceutical University, Guangzhou 510006, China. cxianfa@126.com
- Publication Type:Journal Article
- MeSH:
Algorithms;
Discriminant Analysis;
Gene Expression Profiling;
classification;
methods;
Humans;
Linear Models;
Neoplasms;
genetics
- From:
Journal of Biomedical Engineering
2011;28(6):1213-1216
- CountryChina
- Language:Chinese
-
Abstract:
With its high dimensionalities, small samples and great noise, feature reduction of gene expression profile becomes quite necessary. The most common form of gene expression profile is nonlinear, and traditional dimensionality reduction methods can not project high dimensional data, whose initial dimensionalities are low, into low dimensional space. In this work, an improved distance locally linear embedding (LLE ) algorism was proposed to reduce the dimensionalities. LLE method is very sensitive to the closely-neighboring parameters. In order to enhance the robustness to the number of neighbors, in the paper we presented a novel distance to measure the distance between the samples for the purpose of reducing-the influence of distribution of samples. Experimental results demonstrated that the improved distance LLE can effectively extract information of classification features and greatly reduce the dimensionalities of data while maintaining a higher classification accuracy.
