Concept normalization in free-form texts is a crucial step in every\ntext-mining pipeline. Neural architectures based on Bidirectional Encoder\nRepresentations from Transformers (BERT) have achieved state-of-the-art results\nin the biomedical domain. In the context of drug discovery and development,\nclinical trials are necessary to establish the efficacy and safety of drugs. We\ninvestigate the effectiveness of transferring concept normalization from the\ngeneral biomedical domain to the clinical trials domain in a zero-shot setting\nwith an absence of labeled data. We propose a simple and effective two-stage\nneural approach based on fine-tuned BERT architectures. In the first stage, we\ntrain a metric learning model that optimizes relative similarity of mentions\nand concepts via triplet loss. The model is trained on available labeled\ncorpora of scientific abstracts to obtain vector embeddings of concept names\nand entity mentions from texts. In the second stage, we find the closest\nconcept name representation in an embedding space to a given clinical mention.\nWe evaluated several models, including state-of-the-art architectures, on a\ndataset of abstracts and a real-world dataset of trial records with\ninterventions and conditions mapped to drug and disease terminologies.\nExtensive experiments validate the effectiveness of our approach in knowledge\ntransfer from the scientific literature to clinical trials.\n