Agilent Medip And Mat Algorithm For Enrichment

Question

[kanwarjag]

I have agilent MEdip data and am planning to use MAT algorithm for calling enriched regions. I have seen Cistrome and other publication where they have used MAT on agilent data but I could not find any paper or else where MAT can also be used for Agilent, But for agilent mostly Batman has been used. Has any one used MAT for calling methylation enriched regions. or any reading material if i have missed using MAT fr Agilent MEdip

Thanks

Answer 1

[Charles Warden]

I've used MAT for MeDIP data (specifically MIRA data on NimbleGen tiling arrays), as implemented in Partek Genomics Suite.

Not sure how different the probe density is for Agilent arrays. Either way, you can test multiple algorithms and see what results look best. I'd recommend trying to visualize the log2 ratio between the MeDIP and INPUT signals to verify the results. Total number of candidates is also worth considering: you probably want something in the range of a few hundred peaks.

[kanwarjag]

Thanks cwarden45. What will be best way of visulaizing log2 ratio. Kind of histogram, PCA or else? Sorry about a very navie question

[Charles Warden]

Depends upon the program that you use. Getting the data in the form of a .wig file to view in the UCSC Genome Browser or IGV would probably be the easiest, but these programs accept a variety of formats

http://genome.ucsc.edu/goldenPath/help/wiggle.html

http://www.broadinstitute.org/igv/WIG

If you don't have the option to export such a file (and creating a .wig file seems too difficult), you can check a few regions manually (as long as you can view your normalized signal in table with genomic coordinates).

If that doesn't work, hopefully you will at least have ways to visualize your data with each program you test. For example, I believe Partek has a genome browser: once you open your region of interest, you can view the signal in any specified region.

Alternatively, in your results table (ideally, with both significant and non-significant values), you can try to find the closest match to qualitatively see if each program processed the region in the same way. Not ideal, but it still might be able to help.