Calculating Tissue-Specificity From Affymetrix

Question

[Yang]

Hi everyone,

I'm new to affymetrix data and really need some help for my research. I have .CEL datasets from different tissues and want to calculate tissue-specificity for every gene. After reading some tutorials on bioconductor, I've know the pipeline of dealing such data, but still have some questions on details:

1) The probesets of low-quality need to be removed before calculating gene signal. Is there any packages or any standard method in Bioconductor can do this?

2) For the tissue-specificity of every gene is the goal of my work, I want to know if the gene is expressed or not in a certain sample. So how do I determine the cut-off for expression? Besides, for multiple probesets might map to one gene, how can I integrated the information from different probesets?

I've searched the keywords but got no clear solution. It will be very appreciated if anyone can help. :)

Answer 1

[Nathan Harmston]

I tried to answer some related questions previously:

• Here; and
• Here

Maybe these will help. Determining a cut-off for expression tends to be an arbitrary decision.....sometimes just what gives the best looking volcano-plots (slightly cynical). Maybe some kind of biclustering might help you here for determining the tissue specificity. A quick rseek search shows that there are plenty of implementations of this kind of method. (Biclustering of Expression Data, Cheng, Church). The clusters you get might correspond to tissues, but then again they may not.

HTH a little at the least.

[Khader Shameer]

Another related question is here.

[Yang]

Yeah...it seems like there isn't universally accepted cut-off for "expressed". I had thought it would be like the "10 reads criterion" for RNA-seq data. Now it looks like things are not so easy. Maybe I should have a look at the volcano-plots, as you suggested.

Answer 2

[User 59]

For doing probeset filtering you almost certainly want to look at the genefilter package in BioConductor, have a look at nsFilter. You can also use the all.present and all.present.in.group functions in simpleaffy.

[Yang]

I've done the filtering job by nsFilter based on variability. Thank you!

Answer 3

[Khader Shameer]

AFAIK, several datasets are already available for tissue specificity of microarray probes. If you want to process your raw data files which are not yet submitted in GEO or other public repositories, you may go ahead with suggestion by Nathan and Daniel. In case if you are looking at available data, I would recommend you take a look at BioGPS or HPRD.

BioGPS Gene Expression activity data will be a good start. Another option will be to map the probes that you are interested in, get their NCBI ID and retrieve 'Site of Expression' information from the HPRD flat files. HPRD data is provided from the point of gene and not as such by the affy probes, because of that you need to do the mapping. For each gene-tissue associations curated literature evidence is provided.

[Yang]

I've tried both BioGPS and HPRD, which are quite handy and intuitive. However, both of them are for human only, do you know any datasets for fly? :)

[Khader Shameer]

FlyBase, IMHO one of the best model organism database. See the tab "Expression Data" here. Tissue specific information is available in the 'FlyBase Gene Expression Report' (Example) and under the 'Microarray Data'. Please contact FlyBase curators for raw data files of these datasets.

[Assa Yeroslaviz]

But is it also for miRNA?

Answer 4

[Michael Imbeault]

For the gene cut-off, if you're lucky enough to have A- sexual chromosomes probes and B- samples from females, just pick the average signal of the Y chromosome probe on those samples, which should be background noise.