Improving the learning of chemical-protein interactions from literature using transfer learning and specialized word embeddings

• DOI: 10.1093/database/bay066
• Publication Date: 2018-06-12T19:09:32Z
• Authors: P Corbett; J Boyle
• Keywords: Transfer of learning

In this paper, we explore the application of artificial neural network ('deep learning') methods to problem detecting chemical-protein interactions in PubMed abstracts. We present here a system using multiple Long Short Term Memory layers analyse candidate interactions, determine whether there is relation and which type. A particular feature our use unlabelled data, both pre-train word embeddings also LSTM network. On BioCreative VI CHEMPROT test corpus, achieves an F score 61.51% (56.10% precision, 67.84% recall).