Daniel:Notebook/Deconvolution

From ZhangLabWiki
Revision as of 21:25, 21 November 2013 by >Djacobse (→‎Project)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Back to Notebook <calendar> name=Daniel:Notebook/Deconvolution format=%name/%year-%month-%day date=2013/02/01 view=oneyear </calendar>

Project[edit]

The project is based on the paper by Houseman et al

This project was given to me around January or February of 2013. I worked on it during winter quarter and got part of the way, but never finished it. With my schedule more free I hope to continue. Noi has also been assigned the project recently, so I will be working with her on trying to develop the best ways to do everything.

The project focuses on deconvolution of mixed samples of methylation data, i.e. methylation data from whole blood samples. There are many different types of cells in blood, and each has its own unique DNA methylation. The data from a population of cells, therefore, will be a mix of the individual cell methylation signatures, weighted by the fraction of cells of each type present. To put it mathematically, for given cpg methylation data in a mixed sample beta(i), the beta value will be the weighted sum over all cell types j (from 1 to N types), weighted by the fraction of those cells present, f sub j:

File:CellSig.png

In matrix terms (which will be easier to use in MATLAB):

File:MatrixCellSig.png

Where beta(mixed) is a mx1 vector where m is the number of cpg sites viewed. Beta(pure) is a mxn matrix, where n is the number of cell types. f is a nx1 vector of cell fractions in the mixed sample.

Unfortunately, there are complicating factors:

  • Not all methylation values are useful in determining cell type. Important cpg sites must be located using ANOVA, which can highlight the largest source of variance in a data sample
  • The method requires previous knowledge obtained from homogeneous data. In order for the mixed samples to be effectively analyzed, sufficiently high numbers of homogeneous (pure) samples must be analyzed
  • One thing I found during previous work was that different cells have different distributions of differentially methylated cpgs. Taking, for instance, the 500 most significant cpg sites will yield over 100 differentially methylated sites for cell type A, but only 1 or 2 for cell type B. I will check with Dr. Zhang and/or Athurva, but I think the best way to do the analysis is using equal numbers of significant sites for each cell type