Noi/NOTES/2012-5-12: Difference between revisions
Jump to navigation
Jump to search
>Noi No edit summary |
>Noi No edit summary |
||
Line 26: | Line 26: | ||
4 6.0525615 2.0565664 4.4626373 young adult | 4 6.0525615 2.0565664 4.4626373 young adult | ||
* [[Media:lda_age_10FDR.txt| lda_age_10FDR.R]], [[Media:lda_sex_10FDR.txt| lda_sex_10FDR.R]], [[Media:lda_status_10FDR.txt| lda_status_10FDR.R]] | * [[Media:lda_age_10FDR.txt| lda_age_10FDR.R]], [[Media:lda_sex_10FDR.txt| lda_sex_10FDR.R]], [[Media:lda_status_10FDR.txt| lda_status_10FDR.R]] | ||
[[File:plot_LD1_age_10FDR_test.png]] [[File:plot_LD1_age_10FDR_train.png]] [[File:plot_LD1_sex_10FDR_test.png]] [[File:plot_LD1_sex_10FDR_train.png]] [[File:plot_LD1_status_p.015_test.png]] [[File:plot_LD1_status_p.015_train.png]] | [[File:plot_LD1_age_10FDR_test.png| 400px]] [[File:plot_LD1_age_10FDR_train.png| 400px]] [[File:plot_LD1_sex_10FDR_test.png]] [[File:plot_LD1_sex_10FDR_train.png]] [[File:plot_LD1_status_p.015_test.png]] [[File:plot_LD1_status_p.015_train.png]] |
Revision as of 14:13, 13 May 2012
Regression analysis of UCLA SZ data, applying LDA function
- Links
- I repeated regression analysis again by changing some criteria to select the candidate significant CpG sites, based on FDR or with raw p-value cutoff. For regression on status, I will not apply multiple test correction since I didn't get any significant signals. However with minimum t statistic of 2.5 (p-value ~0.014) from the previous analysis I could separate out the normal and affected samples. Dr. Zhang suggested that this should be fine. In addition, I will change the color of the control samples in PCA analysis plots. LDA will be used then the number of samples in the two groups classified by the line will be counted.
PCA analysis, plot the data in scatter plot
File:UCLA-GA age 10FDR pcaplot1.png File:UCLA-GA age 10FDR pcaplot2.png File:UCLA-GA sex 10FDR pcaplot1.png File:UCLA-GA sex 10FDR pcaplot2.png File:UCLA-GA status pval0.015 pcaplot1.png File:UCLA-GA status pval0.015 pcaplot2.png
- I generate PC score (PC1-PC3). The reason is that the three components seem to contribute to largest variance and the slope change in the scree plot occurs at component 3 (age) as the picture below. Actually, I should try to analyze with different number of component and see if the classification change or not. Note the plot of status seem to be different from age because the slope change occur at component4 or 5.
A=read.table("10%FDR_UCLA-GA_age_methylMatrix.txt",header=TRUE,row.names=1) B=na.omit(A) B$STDEV=NULL B$min_RD=NULL B$mean_RD=NULL pca1 <- prcomp(t(B), scale=TRUE) summary(pca1) v<-data.frame(pca1$x[,1],pca1$x[,2],pca1$x[,3]) names(v)<-c("PC1","PC2","PC3")
- Then add the labels (age, sex or disease status) to each sample in the last column and used this data as the input for LDA
V1 V2 V3 V4 1 0.8641308 3.1103773 -1.2768520 adult 2 -7.3738694 -3.7825118 2.4297650 adult 3 6.3925767 0.5246122 0.9405213 young adult 4 6.0525615 2.0565664 4.4626373 young adult
File:Plot LD1 age 10FDR test.png File:Plot LD1 age 10FDR train.png File:Plot LD1 sex 10FDR test.png File:Plot LD1 sex 10FDR train.png File:Plot LD1 status p.015 test.png File:Plot LD1 status p.015 train.png