Constructing a Gene-Drug-Adverse Reactions Network and Inferring Potential Gene-Adverse Reactions Associations Using a Text Mining Approach

Our objective was to identify and extract gene-drug and drug-adverse drug reaction (ADR) relationships from different biomedical literature collections, and to predict the possible association between gene and ADR. The drug, ADR and gene entities were recognized by a CRF model with multiple features. Logistic regression models were constructed for each drugADR and drug-gene pair based on its frequency, Mesh Rule association and similarity with known association etc. Using predicted score to generate drug- ADR matrix and drug-gene matrix, and then calculating for gene-ADR matrix. Network and clustering analysis were applied to verify and interpret the relationship between them. A total of 78014 potential gene ADR associations were predicted. Part of the predicted results can be explained by the network-clustering-pathway analysis, and verified in the literature. The gene-drug-ADR network constructed in this study can provide a reference for the possible association between the gene and ADR.