Multi-type Microbial Relation Extraction by Transfer Learning

Microbial interaction network is the foundation to understand the structure and function of microbial communities. However, there is currently less a comprehensive dataset of microbial interaction network. Using text mining technology, available microbial interaction knowledges could be extracted automatically from unstructured biomedical text data. However, the existing biomedical relation extraction tasks can only identify binary relations among microorganisms without differentiating complex interaction types. In this paper, we proposes a computational framework for the task of multi-type microbial relation extraction based on transfer learning. Transfer learning models were applied on large-scale unlabeled texts from PubMed, and predicted 2,132 standardized multi-type microbial interaction relationships among 682 bacterial species.