Since co-expressed genes often are co-regulated by a group of transcription factors, different conditions (e.g. disease versus normal) may lead to different transcription factor activities and therefore different co-expression networks. We propose a method for identifying condition-specific co-expression networks by combining our recently developed network quasi-clique mining algorithm and the expected conditional F-statistic. We apply this method to compare the transcriptional programmes between the non-basal and basal types of breast cancers. The results provide a new perspective for studying gene interaction dynamics in cancers and assessing the effects of perturbation on key genes such as transcription factors. Our work is a way for dynamically characterising the gene interaction networks.