Dinh/Dinh 2012/NOTES/2012-2-21
Jump to navigation
Jump to search
48 African Samples
- Cleaning with Tech_Rep_9 data: Using requirement of (observed STDEV > 0.1 and p-value < 0.05) I obtained a list of 107,218 bad_cpgs (out of 412,476). The positions are 1-based and on Hg19.
- Next, I filtered out the sites using this script: File:Getsites.txt (This should be used temporarily, we need a more standardized script). The current UPENN methylMatrix were generated using Dr. Zhang's script so it has 0-based positions.
- wc -l:
56092 cleanUPENN46_hg19_methylMatrix.txt 77573 UPenn46_hg19_methylMatrix_filtered.txt
Comparing before and after cleaning
- Before cleaning:
> A = read.table("UPenn46_hg19_methylMatrix_filtered.txt", header=TRUE, row.names=1) > B = na.omit(A) > str(B) 'data.frame': 53548 obs. of 46 variables: ... > B$STDEV=NULL > B$min_RD=NULL > B$mean_RD=NULL > pca_sites = prcomp(B, scale=TRUE) > summary(pca_sites) ... PC1 PC2 PC3 PC4 PC5 PC6 PC7 Proportion of Variance 0.723 0.0376 0.0125 0.0105 0.00979 0.00869 0.0076 ... > write.table(file="pca_sites.orig.rotations", pca_sites$rotation) > pca_indiv = prcomp(t(B), scale=TRUE) > summary(pca_indiv) ... PC1 PC2 PC3 PC4 PC5 PC6 PC7 Proportion of Variance 0.205 0.0382 0.0376 0.0336 0.0327 0.0230 0.0216 ... > write.table(file="pca_indiv.orig.x", pca_indiv$x)
- After cleaning:
> A = read.table("cleanUPENN46_hg19_methylMatrix.txt", header=TRUE, row.names=1) > C = na.omit(A) > C$STDEV=NULL > C$min_RD=NULL > C$mean_RD=NULL > str(C) ... 'data.frame': 37135 obs. of 46 variables: ... > pca_sites = prcomp(C, scale=TRUE) > summary(pca_sites) ... PC1 PC2 PC3 PC4 PC5 PC6 PC7 Proportion of Variance 0.721 0.0394 0.0125 0.0106 0.0092 0.00861 0.00761
> write.table(file="pca_sites.clean.rotations", pca_sites$rotation) > pca_indiv = prcomp(t(C), scale=TRUE) > summary(pca_indiv) ... PC1 PC2 PC3 PC4 PC5 PC6 PC7 Proportion of Variance 0.212 0.0378 0.0366 0.0332 0.030 0.0231 0.0217 > write.table(file="pca_indiv.clean.x", pca_indiv$x)
Results:
PCA_sites: calculate the correlation of the rotation (weight) of each samples onto CpG sites with QC factor PC1 PC2 PC3 Original data: %variance 72.3 3.76 1.25 Pearson's cor(R) 0.77 0.73 0.10 Clean data: %variance 72.1 3.94 1.25 Pearson's cor(R) 0.77 0.72 0.07
- There are barely any change after cleaning on PCA_sites. As hypothesized before, PCA sites show that higher quality samples tend to contribute more variation between the CpG sites, so if we remove poor quality sites, it will not affect this measure.
PCA_indivs: calculate the correlation of the rotated value of each samples from linear combination of CpG sites with QC factor PC1 PC2 PC3 Original data: %variance 20.5 3.82 3.76 Pearson's cor(R) -0.74 -0.23 -0.45 Clean data: %variance 21.2 3.78 3.66 Pearson's cor(R) -0.74 0.53 -0.03
- The correlation on PC1 is the same, but PC2 becomes positively correlated while the correlation on PC3 dropped to insignificant.