negative geo value analysis

Question

[linouhao]

hi,

GSE55542 is a geo data, for example, one chip in it

https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM1338936

it has negative value, so can I use limma to analyse this data?

thanks a lot

Answer 1

[Ahill]

The annotation on this GSM says the reported expression value is:

Normalized signal intensity (Background corrected, log2 transformed, quantile normalized and base line transformed using the median of all samples).

Since it is a log2-transformed there is nothing intrinsically 'wrong' about having negative values, and it could be suitable for linear modelling with limma. But you should definitely do standard QC steps on the GSE series data (e.g. boxplots, PCA, correlation plots, M-A plots) to assess the normalization and distribution of the expression values before doing differential analysis.

[linouhao]

thanks a lot, I do not have much experience here, and can not handle the normalization well, either normalwithinarrays and normalbetweenarrays. the negative values here maybe not because of log, but also of backgroud substraction, because many gse also do rma and log, but no negative values. do you have time to share your experience about this concrete gse?