Deep learning for extracting protein-protein interactions from biomedical literature

摘要

State-of-the-art methods for protein-protein interaction (PPI) extraction areprimarily feature-based or kernel-based by leveraging lexical and syntacticinformation. But how to incorporate such knowledge in the recent deep learningmethods remains an open question. In this paper, we propose a multichanneldependency-based convolutional neural network model (McDepCNN). It applies onechannel to the embedding vector of each word in the sentence, and anotherchannel to the embedding vector of the head of the corresponding word.Therefore, the model can use richer information obtained from differentchannels. Experiments on two public benchmarking datasets, AIMed and BioInfer,demonstrate that McDepCNN compares favorably to the state-of-the-artrich-feature and single-kernel based methods. In addition, McDepCNN achieves24.4% relative improvement in F1-score over the state-of-the-art methods oncross-corpus evaluation and 12% improvement in F1-score over kernel-basedmethods on "difficult" instances. These results suggest that McDepCNNgeneralizes more easily over different corpora, and is capable of capturinglong distance features in the sentences.