文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > The Analysis of Microarray Datasets Using a Genetic Programming
侵权投诉

The Analysis of Microarray Datasets Using a Genetic Programming

The Analysis of Microarray Datasets Using a Genetic Programming

MEF2C is activated by NKX2-5 at both the RNA and protein levels. Its overexpression inhibits expression of NR4A1, which regulates apoptosis via BCL2 transformation [19]. Some other genes are AML associated, such as IL18, CCL5, SPI1, FUT4 and MPO. Biological experiments have demonstrated their behaviors in AML development. The heat maps of expressions of the top30 genes in both leukemia datasets are shown in Fig 2. (E), (F). Except the null gene, PTARF and FOXF2, the other genes are almost the same when clustering them into two groups according to their expressions in all samples.

For the breast cancer dataset, we analyzed the top30 genes in NCBI. Some of these genes are closely relevant with the breast cancer, such as MMP11, CAPG and so on. Data confirmed MMP's contribution to breast cancer progression and its expression was shown to have a direct negative effect on patients' survival [20]. For the gene CAPG, it has been reported that its mobility in cell nuclei of live breast cancer cells [21]. Such evidences showed the meaningful results of our method.

We also checked our results in terms of the pathways in which the top30 genes participate. For the prostate cancer datasets, since there are only 850 common genes of the two datasets. 30 genes with the highest z-scores are analyzed of their participating pathways with the tool of Onto-Express [22], [23]. The most probable pathways with these 30 genes are shown in the following table. It is found that the prostate cancer pathway is ranked six of all the probable pathways. During the 850 common genes, there are 5 genes participating in the prostate cancer genesis, where 89 genes in all participate in this process. Other pathways are very important in the oncogenesis, such as Hedgehog signaling pathway and MAPK signaling pathway.

TABLE Ⅱ. THE MOST SIGNIFICANT PATHWAYS OF THE TOP30 GENES IN PROSTATE CANCER DATASET

Rank Pathway

Name

Pathway genes in the

common genes

1 Dentatorubropallidoluysian 15 1 2

atrophy

2 Hedgehog signaling 57 1 3

pathway

3 MAPK signaling pathway 265 2 15 4 Regulation of actin 211 2 16

cytoskeleton

5 44 1 4 TypeⅡ diabetes mellitus 6 Prostate caner 89 1 5 7 Long-term potentiation 69 1 6 8 Glycan 122 1 8

structures-biosynthesis 1

9 GnRH signaling pathway 97 1 8 10 Calcium signaling pathway 175 1 8

Genes in

the pathway

Input genes

p-value

are ranked high. In the breast cancer dataset, some signaling pathways which are involved in the genesis of a series of cancers exist in the pathways analyses, for example, p53 signaling pathway, Wnt signaling pathway and MAPK signaling pathway.

In each run, we selected 50 best classifiers and obtained 1000 classifiers in total. The accuracy of each classifier in all samples has been counted. In prostate datasets, the best classifier can achieve 97.3% correct classification of training samples. That is, only 2 of 74 training samples are classified incorrectly. This classifier can achieve the accuracy of 86.84% in test samples and 100% in independent samples. In leukemia datasets, the best classifier correctly classifies all the training and test sets as well as 30 samples in 37 independent test set. The experiment design and best classifier accuracy are shown in Table 3.

TABLE Ⅲ. THE EXPERIMENT DESIGN AND BEST CLASSIFIER ACCURACY.

Prostate cancer datasets ClassifierAccuracyLeukemia datasets ClassifierAccuracyBreast cancer datasets

Classes

Training setTest set

Tumor4724Normal 2714adjacent tissue

if plus(X516,min(X132,max(minus(X132,X177),X569)))>0.5931, then nornal prostate ; else prostate cancer

97.3%86.84%100%

ALL331721

AML2411 16 if gt(X3665,X2515), then ALL; else AML100%100%81.08%Tumor 3110

16 7 5

9

Independent

test set144

Classifier Accuracy

Normal tissue

if plus(X1366,max(X2656,max(X1556,X906)))<1.2828, then breast cancer ; else normal breast

92.9% 90.5% 100% IV. CONCLUSIONS

In this paper, we utilized the GP to multiple cancer

datasets. The advantage of GP is that through wide search of genes and continuous transmutations, an amount of classifiers that have good performances in recognition of

0.09

diseases could be found. The mathematic expressions of

0.13 selected genes may reveal the underlying relation of them in

the genesis of diseases. Another contribution for this paper is 0.15

0.17 to compound different datasets obtained in different labs. 0.17 This method could do it best to alleviate the systematic and 0.21 random noises inherent to respective dataset. The obtained 0.25 classifiers have more generalization ability in classification 0.32

and better representation of essence of diseases. In spite of

0.32 this advantage, it may exclude a number of instructive genes 0.32

that don’t belong to intersection of probes in cross datasets.

However, this problem could be avoided by choosing For the leukemia datasets and breast cancer datasets, since

appropriate datasets that have a large coincidence of the gene

there are more than 4000 common genes between them, we

pools. The result proves that our method is valid in seeking

checked 100 genes with the highest z-scores in terms of their

classifiers of common characteristics of diseases.

pathways with the help of Onto-Express. In the leukemia

datasets, some important pathways, such as leukocyte transendothelial migration and MAPK signaling pathway,

上一页第5页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email