DRExplainer: Quantifiable Interpretability in Drug Response Prediction with Directed Graph Convolutional Network

TL;DR

This work tackles directional drug response prediction in cancer by introducing DRExplainer, a directed bipartite graph convolutional network that fuses multi-omics cell-line data with drug molecular graphs. A DistMult decoder enables robust link scoring, while a novel explainer learns a mask over the directed adjacency to produce quantitatively evaluable, subgraph-based explanations, benchmarked against ground-truth biological knowledge. The model demonstrates superior predictive performance (AUC/AUPR) over state-of-the-art baselines and offers interpretable insights validated by case studies, supporting its potential clinical and preclinical utility. Additionally, a ground-truth interpretability benchmark and precision/recall-based metrics enable rigorous assessment of explanations, advancing trustworthy application of graph-based methods in precision oncology.

Abstract

Predicting the response of a cancer cell line to a therapeutic drug is pivotal for personalized medicine. Despite numerous deep learning methods that have been developed for drug response prediction, integrating diverse information about biological entities and predicting the directional response remain major challenges. Here, we propose a novel interpretable predictive model, DRExplainer, which leverages a directed graph convolutional network to enhance the prediction in a directed bipartite network framework. DRExplainer constructs a directed bipartite network integrating multi-omics profiles of cell lines, the chemical structure of drugs and known drug response to achieve directed prediction. Then, DRExplainer identifies the most relevant subgraph to each prediction in this directed bipartite network by learning a mask, facilitating critical medical decision-making. Additionally, we introduce a quantifiable method for model interpretability that leverages a ground truth benchmark dataset curated from biological features. In computational experiments, DRExplainer outperforms state-of-the-art predictive methods and another graph-based explanation method under the same experimental setting. Finally, the case studies further validate the interpretability and the effectiveness of DRExplainer in predictive novel drug response. Our code is available at: https://github.com/vshy-dream/DRExplainer.

Figures (6)

• Figure 1: Overview of DRExplainer framework. (A) DNNs and GNNs separately encode the input SMILES and multi-omics profiles. The encoded drug and cell line representations are fed into our directed graph convolutional network to learn further node embeddings. The updated embeddings are decoded by the DistMult decoder to predict novel responses. After prediction, the explainer and evaluator modules are employed to explain the prediction and evaluate the explanation, respectively. (B) The overview of explainer architecture. (C) The overview of evaluator architecture
• Figure 2: Three ground truth construction methodologies for drug response prediction.
• Figure 3: The AUC and AUPR scores of DRExplainer in inductive capability learning.
• Figure 4: The receiver operating characteristic and precision-recall curve of DRExplainer and baselines on the independent test.
• Figure 5: The performance of DRExplainer and baselines on the cross-validation.