I'd be extremely hesitant to call differentially expressed genes based on completely different sets of data. RNA-seq directly sequences cDNA, microarrays measure the fluorescence emitted by the labeled target sequences to probe sequences spotted on the chip. They come with very different sources of bias, not to mention the batch effects of having different persons handle the cDNA extracted from different samples at different places at different times with different experimental protocols...The whole point of the control data set for calling DE genes is to estimate the baseline/background expression values. By cooking up a control that doesn't match any of the expected characteristics of your treatment sample, I don't think you're doing yourself a favor.
That being said, I don't completely understand why an appropriate control sample is missing from the original submission. GSE7562 gives plenty of replicates for PTEN loss as well as WT. (found by following GPL570 which was indicated as the original data set in the accession number you mentioned)
Is this data downloaded from public repositories like (NCBI GEO, SRA (or) Array express). In that case the control and case samples are generated at different time and conditions, so there is a high chance of batch effect. First, batch effect need to be corrected.
DO you have biological replicates for both case and control ?