Theoretically yes, but your problem is that you do not know what ratio of fibroblasts to cardiomyocytes you have in your sample, and that their contribution to each expression signal (for instance) will not be linear by the nature of microarrays.

I think your best bet is to do the experiment again without fibroblasts. You could simply remove the genes that fibroblasts are known to express, however you do not know whether these are important in your cardiomyocytes.

Hi Andrew , your completely right, your suggestion was also my first thought. However redo the whole experiment is something else. So i was looking for alternatives to save my experimental output.

There are R/Matlab packages that can deconvolute mixed microarray data (i.e. OncoSNP for SNP/CNV arrays), I wonder if you could adapt these to solve your problem.

Essentially it is one of heterogeneity and if you can build a Hidden Markov Model (HMM) to model your dataset you should be able to correct for the effect of the fibroblast contamination as it is essentially like "normal" tissue contamination. i.e. something like this but for expression arrays:

http://www.biomedcentral.com/1471-2105/14/S5/S11

I would recommend talking to a good bioinformatician.