Kun:LabNotes/CpgSeq/2012-2-6
Jump to navigation
Jump to search
Second analysis of BSPP data from Geographic Astrophy (GA) patients[edit]
- This is the continuation of the first analysis did on 2011-10-14.
- The purpose of this analysis is to use the UCLA SZ samples as the controls to test the accuracy of classification based on methylation markers.
- Noi did the mapping and generated BED files for both data sets. UCLA SZ data; UCSD GA data
- I generated the methylation matrix by myself using my allBED2Matrix.pl script, because there is a 1bp difference in CpG positions between my script and Dinh's version.
- For the GA samples, I required that there are at least 58 methylation values at each CpG site, and the standard deviation is at least 0.1.
- For the UCLA SZ samples, I required at least 50 methylation values, but no threshold for standard deviation, because the informative CpG sites identified from the GA samples might not be variable in the SZ samples.
- I made one modification in the GA_methylation_MRMR_prep.pl, so that all candidate CpG sites have no missing value.
./GA_methylation_MRMR_prep.pl Feb2012/GA_goodqual_min58_minSTD0.1_methylMatrix_noSNP.txt > Feb2012/GA_goodqual_min58_minSTD0.1_methylMatrix_noSNP_MRMR_input.csv & ~kunzhang/softwares/mrmr_c_src/mrmr -i GA_goodqual_min58_minSTD0.1_methylMatrix_noSNP_MRMR_input.csv > GA_goodqual_min58_minSTD0.1_methylMatrix_noSNP_MRMR_output.txt
- I took the top 50 MaxRel features, and extract the methylation values on these sites from the UCLA SZ data set.
../extract_N_prep_matrix.pl GA_goodqual_methyl_min58_minSTD0.1.noSNP.50_marker_list UCLA-4batches_all_methylMatrix.txt
- I then manually removed the CpG sites (or samples) that have missing values, then combined the UCLA SZ data and UCSD GA data on 35 CpG sites: UCLA-SZ-UCSD_GA_35_marker.methylMatrix.
- PCA analysis was performed on this combined matrix. Then I plotted all 166 samples based on the first two principal components.
Green: GA controls; Red: GA cases; Blue: SZ samples (unrelated controls) File:PCA classification 35 cpg sites.png
- Conclusion: Clearly the SZ samples cluster closely to the GA controls than the GA cases, which is a good indication that these methylation markers (as well as this strategy) could be valid.