IMSE: interaction information attention and molecular structure based drug drug interaction extraction

Area of Science:

• Biomedical Informatics
• Computational Chemistry

Background:

• Drug drug interaction (DDI) extraction is crucial for drug safety monitoring.
• Current methods using BiLSTM and BERT lack optimal feature representation.

Purpose of the Study:

• To develop a novel DDI prediction model with enhanced feature representation.
• To improve the accuracy of identifying potential adverse drug reactions.

Main Methods:

• Integration of an attention vector to handle overlapping relations.
• Incorporation of drug molecular structure information.
• Utilization of text features combined with T-distribution and chi-square distribution.

Main Results:

• Achieved similar or superior prediction performance compared to state-of-the-art DDI models.
• Obtained F-scores up to 85.16% on benchmark datasets.
• Demonstrated improved focus on drug entities through enhanced text features.

Conclusions:

• The proposed model, leveraging transformer architecture and multiple features, enhances DDI prediction performance.
• This approach is valuable for identifying potential adverse drug reactions in the biomedical domain.