Automating ontology curation is a crucial task in knowledge engineering. Prediction by machine learning techniques such as semantic embedding is a promising direction, but the relevant research is still preliminary. In this paper, we present a class subsumption prediction method named BERTSubs, which uses the pre-trained language model BERT to compute contextual embeddings of the class labels and customized input templates to incorporate contexts of surrounding classes. The evaluation on two large-scale real-world ontologies has shown its better performance than the state-of-the-art.