ExDDI: Explaining Drug-Drug Interaction Predictions with Natural Language

TL;DR

This work proposes to generate natural language explanations for DDI predictions, enabling the model to reveal the underlying pharmacodynamics and pharmacokinetics mechanisms simultaneously as making the prediction.

Abstract

Predicting unknown drug-drug interactions (DDIs) is crucial for improving medication safety. Previous efforts in DDI prediction have typically focused on binary classification or predicting DDI categories, with the absence of explanatory insights that could enhance trust in these predictions. In this work, we propose to generate natural language explanations for DDI predictions, enabling the model to reveal the underlying pharmacodynamics and pharmacokinetics mechanisms simultaneously as making the prediction. To do this, we have collected DDI explanations from DDInter and DrugBank and developed various models for extensive experiments and analysis. Our models can provide accurate explanations for unknown DDIs between known drugs. This paper contributes new tools to the field of DDI prediction and lays a solid foundation for further research on generating explanations for DDI predictions.

Figures (7)

• Figure 1: Examples of different DDI prediction tasks. (a) Model inputs, i.e., SMILES representations of the drug pairs; (b) Traditional DDI prediction: binary classification task; (c) DDI-type prediction: multiple classification task; (d) DDI explanation generation: our proposed task, formulated as text generation. The underlined content represents the annotations involved in DDI type prediction, while the italicized text denotes unique content provided by DDInter's explanations.
• Figure 2: Illustration of the fine-tuning methods. ① the learning objective of the ExDDI-S2S model; ② the learning objective of the ExDDI-MT model; ③ the inference step of the ExDDI-MTS model.
• Figure 3: Binary classification results. Green and red dashed lines indicate results from the GMPNN-CS and DSN-DDI papers, respectively. Gold and orange bars represent models trained with DrugBank or DDInter explanations. Mean values over 5-fold cross-validation are shown for all models except ExDDI-IC. The error bars represent the standard deviation. Detailed numbers with precision and recall scores are available in the Appendix D.
• Figure 4: Multiple classification results. Green and red dashed lines indicate results from the DDIMDL and NMDADNN papers, respectively. Gold and orange bars represent models trained with DrugBank or DDInter explanations. Mean values over 5-fold cross-validation are shown for all models except ExDDI-IC. The error bars represent the standard deviation. Detailed numbers with precision and recall scores are available in the Appendix D.
• Figure A1: Explanation generation results. Mean values and standard deviations from 5-fold cross-validation are presented for all models except ExDDI-IC, which was run only once due to the high cost of API calls and its low performance. The best results for each dataset are highlighted in bold.