Dinh/Dinh 2012/NOTES/2012-1-27

From ZhangLabWiki
Jump to navigation Jump to search

March 16 additions[edit]

  • Technical + sampling variability of BSPP data:
Correlate technical standard deviation and p-values with probes efficiencies.
So we can remove technical variability for every captured CpG (using probes efficiency).
For PCA analysis, we can calculate a subset of sites with low err (Nb/Na) = UPenn, GA

Progress and plan for BSPP dataset[edit]

  • All new data should be mapped to Hg19 - this is important for calling mSNP.
  • We should have a centralized data storage and analysis organization system such that everything can be easily shared.

0. Technical replicates (9, HAPMAP sample) - (Finish in 2 weeks)

 A reference and cleanup script
 Choose a hard or soft cut-off that can help us clean up all the other datasets

1. UCLA (96 pedigrees) - Noi's focus

 Use HAPMAP pedigrees or GA data as control
 mQTL (mSNP) and mQTL (gSNP) => stronger point - can we find mendelian transmission?
 genotype : ASM, mQTL, mPO
 phenotype : disease, age (96, children only) [Linear modeling, classifications]

2. GA (96 case/control)

 Crossvalidation: Use other dataset as more control, can the set of CpG group them into control?

3. Penn (48 no phenotype )

 First we need to do clean up using the technical replicates
 mQTL (mSNP) and mQTL (qSNP)
 Lets do mQTL on all (mSNP) and (gSNP) for all the dataset -- we should see MORE significant sites for UPENN because there is more genetic diversity.

4. N37 (10 tissues and RNA-seq)

 /home/kunzhang/CpgMIP/Data/N37_10_tissues
 How functional is a CpG site, and if it is functional (use a p value of e-3), is it co-localizing with a specific mark?
 Is a functional CpG site also highly tissue specific?
 Obtain sets of CpG that are highly tissue specific to one tissue, just by expression of mC values, can we discern highly tissue specific sites? 
 Exons specific expression regulation
 How many CpG-gene is close range / long range / does it skip over genes? / long stretch of CpG correlate with gene expression
 How many domains can we classify ? How many are ASM domains? 
 add in the average of m-number of blood data => what are the CpG sites that are highly specific to just blood? which are not? then we can use this information for UCLA and AMD data. (average whole blood). and average RNA-seq from whole blood published datasets.

5. HIV (37 sample with phenotype)

 Low priority atm.