we propose a novel FusionGDA model, which utilises a pre-training phase with a fusion module to enrich the gene and disease semantic representations encoded by pre-trained language models.