How To Use Dwd R Package In Order To Remove Biases And Merge Two Microarray Datasets

Question

[fbrundu]

Hi all, I am trying to find a way to use distance weighted discrimination method to remove biases from multiple microarray datasets.

My starting point is this. The problem is that Matlab version runs only under Windows, needs excel 5 format as input (where data appears to be truncated at line 65535 - matlab error is:

Error reading record for cells starting at column 65535.  Try saving as Excel 98.

). Java version runs only with caBIG support, which, if I understood, has been shut down recently.

So I searched a lot and I find this package but from example I could not get how to provide the two datasets to merge to kdwd function.

Does anybody know how to use it?

Thanks

[Michael]

I am not sure that this is the right package. From the documentation it seems that kdwd implements a supervised classifier which is not really what you need for bias removal. Possibly, the classifier uses DWD for preprocessing and feature selection. Unfortunately, I have no clue about the method DWD (Distance Weighted Discrimination) to be able to give further hints.

Answer 1

[Chris Cabanski]

It seems like there are a few different issues, so let me try to address them all separately.

First, I have experience with running DWD in Matlab, but I have never tried the R implementation. I believe that there is a newer version of DWD for Matlab that runs on Mac/unix machines. You can download it from Steve Marron's website. It looks like the main link is down, but you should be able to download the zip folder of all of his functions and find the appropriate Batch Adjust folder and functions.

The excel loading error is separate from DWD. Do you really have over 65,000 columns? If the excel upload isn't working, you can just upload the data from a text file.

If you are still having issues (with the R or Matlab implementation), I would recommend emailing Steve Marron directly. As a former student of his, I know that he is good about replying to his email and helping resolve any issues related to his code.

In response to Michael's comment, DWD is meant to reduce bias that is associated with a known factor, such as the batch number in microarray experiments (assuming that batch is not counfounded with treatment/control). This is different from normalization methods like RMA or quantile normalization, which may not adequately remove batch effects.

[fbrundu]

Thanks for your reply Chris.. I'll try it out in a few days and let you know.. This helps me a lot :)