BactInt: A domain driven transfer learning approach and a corpus for extracting inter-bacterial interactions from biomedical text
This addresses the challenge of efficiently mining microbial interactions from vast biomedical texts for researchers in microbiology and bioinformatics, though it is incremental as it builds on existing extraction methods.
The paper tackles the problem of automatically extracting inter-bacterial interactions from biomedical literature, introducing a transfer learning method and the first publicly available dataset for this task, achieving improved accuracy.
The community of different types of microbes present in a biological niche plays a very important role in functioning of the system. The crosstalk or interactions among the different microbes contributes to the building blocks of such microbial community structures. Evidence reported in biomedical text serves as a reliable source for predicting such interactions. However, going through the vast and ever-increasing volume of biomedical literature is an intimidating and time consuming process. This necessitates development of automated methods capable of accurately extracting bacterial relations reported in biomedical literature. In this paper, we introduce a method for automated extraction of microbial interactions (specifically between bacteria) from biomedical literature along with ways of using transfer learning to improve its accuracy. We also describe a pipeline using which relations among specific bacteria groups can be mined. Additionally, we introduce the first publicly available dataset which can be used to develop bacterial interaction extraction methods.