There are a couple of posts that talk about tackling the problem of mulit-probes targeting a single transcript/gene, and as I found, taking the medium or mean of the expression values would be a good solution. So that's OK, but my problem is that I have some probes that target several transcripts, so that I cannot figure out how to select a transcript to be representative of that probe.
To be more specific, I am working on a microarray probeset with Agilent platform, which is designed to target some human coding and non-coding RNAs. I used BLAST against a local non-coding RNA database, and found that some probes target more than one transcripts (though they are from one single gene)
I should add that my choice to work on "transcripts" level rather than "genes", was due to the fact that there may be a differential expression in one transcript from a gene, but not for other transcripts for the same gene, and so any method to integrate expression values for these transcriptions, may undervalue the importance of that gene.