Chemical-disease relation (CDR) extraction is significantly important tovarious areas of biomedical research and health care. Nowadays, manylarge-scale biomedical knowledge bases (KBs) containing triples about entitypairs and their relations have been built. KBs are important resources forbiomedical relation extraction. However, previous research pays littleattention to prior knowledge. In addition, the dependency tree containsimportant syntactic and semantic information, which helps to improve relationextraction. So how to effectively use it is also worth studying. In this paper,we propose a novel convolutional attention network (CAN) for CDR extraction.Firstly, we extract the shortest dependency path (SDP) between chemical anddisease pairs in a sentence, which includes a sequence of words, dependencydirections, and dependency relation tags. Then the convolution operations areperformed on the SDP to produce deep semantic dependency features. After that,an attention mechanism is employed to learn the importance/weight of eachsemantic dependency vector related to knowledge representations learned fromKBs. Finally, in order to combine dependency information and prior knowledge,the concatenation of weighted semantic dependency representations and knowledgerepresentations is fed to the softmax layer for classification. Experiments onthe BioCreative V CDR dataset show that our method achieves comparableperformance with the state-of-the-art systems, and both dependency informationand prior knowledge play important roles in CDR extraction task.