Transformed count value type in Gene Expression Omnibus(GEO) datasets GDS Series

Question

[khanaparajita]

What is the meaning of "transformed count" value type in NCBI GEO datasets. How is "transformed count" value type different from "log ratio" value type in GDS series of GEO Datasets?

Answer 1

[Devon Ryan]

The meaning of that can vary with every dataset. Similarly, the way in which the log ratio is derived can also vary with every dataset. You'll have to read the rest of the meta information in the dataset to find out how those values were derived. If the information isn't in there, then the only way to find out is by asking the authors.

[khanaparajita]

Log ratio tells how much a gene is regulated with respect to control, 1 means 2 fold over expression, -1 means 2 fold down regulation. Similar to this, what does transformed count refer to? Data set I am using is GDS2771, what does transformed count value corresponding to the genes mean?

[Devon Ryan]

Yes, that's the general concept of a log ratio, but it doesn't tell you how the values used were derived. For GDS2771, those are RMA normalized values (just click over to the series and then on a sample, though this is probably mentioned in the .soft file).

[khanaparajita]

Yes that I got from the soft file, that values are RMA normalized, I am not very clear on what RMA normalized values signify, could you explain. And also can you suggest how can I discretize the gene expression values into 3 levels like ( over expressed, normal expressed and down regulated ) so that the discretized expression levels can be used later for drawing regulatory relationships among the genes. As in log ratio expression values, values say less than -1 can be down regulated, values varying between -1 to 1 are normal expressed and values greater than 1 are over expressed values. Similarly, for transformed count expression values can lower values say in range of 2-5 is down expressed, 5-7 normal expressed and values greater than 7 are over expressed. Can such discretization really capture different expression levels of genes that I wish to capture?

[Devon Ryan]

Just search google or pubmed for RMA normalization.

"over expressed" is a concept that only make sense in reference to something. So just use limma and categorize things according to the adjusted p-value and fold-change.