Reproducibility and Generalization of a Relation Extraction System for Gene-Disease Associations

Biomedical literature is a rich source of information on Gene-Disease Associations (GDAs) that could help physicians in assessing clinical decisions and improve patient care. GDAs are publicly available in databases containing relationships between gene/miRNA expression and related diseases such as specific types of cancer. Most of these resources, such as DisGeNET, miR2Disease and BioXpress, include also manually curated data from publications. Human annotations are expensive and cannot scale to the huge amount of data available in scientific literature (e.g., biomedical abstracts). Therefore, developing automated tools to identify GDAs is getting traction in the community. Such systems employ Relation Extraction (RE) techniques to extract information on gene/microRNA expression in diseases from text. Once an automated text-mining tool has been developed, it can be tested on human annotated data or it can be compared to state-of-the-art systems. In this work we reproduce DEXTER, a system to automatically extract Gene- Disease Associations (GDAs) from biomedical abstracts. The goal is to provide a benchmark for future works regarding Relation Extraction (RE), enabling researchers to test and compare their results. The implemented version of DEXTER is available in the following git repository: https://github.com/mntlra/DEXTER.