Comparing Methods for Drug-Gene Interaction Prediction on the Biomedical Literature Knowledge Graph: Performance vs. Explainability

Abstract

This paper applies different link prediction methods on a knowledge graphgenerated from biomedical literature, with the aim to compare their ability toidentify unknown drug-gene interactions and explain their predictions. Identifyingnovel drug-target interactions is a crucial step in drug discovery and repurposing.One approach to this problem is to predict missing links between drug and genenodes, in a graph that contains relevant biomedical knowledge. Such a knowledgegraph can be extracted from biomedical literature, using text mining tools. In thiswork, we compare state-of-the-art graph embedding approaches and contextualpath analysis on the interaction prediction task. The comparison reveals atrade-off between predictive accuracy and explainability of predictions. Focusingon explainability, we train a decision tree on model predictions and show how itcan aid the understanding of the prediction process. We further test the methodson a drug repurposing task and validate the predicted interactions againstexternal databases, with very encouraging results.