Drug-drug interaction prediction based on co-medication patterns and graph matching

Wen-Hao Chiang, Li Shen, Lang Li, Xia Ning

Abstract

Background: The problem of predicting whether a drug combination of arbitrary orders is likely to induce adverse drug reactions is considered in this manuscript. Methods: Novel kernels over drug combinations of arbitrary orders are developed within support vector machines for the prediction. Graph matching methods are used in the novel kernels to measure the similarities among drug combinations, in which drug co-medication patterns are leveraged to measure single drug similarities. Results: The experimental results on a real-world dataset demonstrated that the new kernels achieve an area under the curve (AUC) value 0.912 for the prediction problem. Conclusions: The new methods with drug co-medication based single drug similarities can accurately predict whether a drug combination is likely to induce adverse drug reactions of interest.

Keywords: drug-drug interaction prediction; drug combination similarity; co-medication; graph matching

Introduction

Most research into Drug-Drug Interactions (DDIs) focuses on pairs—what happens when Drug A meets Drug B. However, the reality of modern medicine, especially for elderly patients, involves "polypharmacy," where individuals often take five or more medications simultaneously. These "high-order" interactions are significantly harder to predict because the number of possible combinations grows exponentially with each added drug.

Existing predictive methods often rely solely on the chemical structures of drugs. While useful, structure does not always capture how a drug behaves within a complex biological system or the clinical context in which it is prescribed. This study addresses these gaps by predicting adverse drug reactions (ADRs) for combinations of arbitrary orders using a combination of structural data and clinical co-occurrence patterns.

Methodology

The researchers formulated the prediction task as a binary classification problem using Support Vector Machines (SVMs). The core innovation lies in the development of a similarity measure between drug combinations of varying sizes to serve as the SVM kernel.

Graph Representation and Matching

Each drug combination is represented as a "complete graph" where nodes represent individual drugs and edges represent the relationships between them. To compare two combinations (e.g., a 3-drug set versus a 5-drug set), the authors utilized Graph Matching Kernels ($K_{gm}$). This approach employs a Linear Sum Assignment Problem (LSAP) algorithm to find the optimal "best fit" between the drugs in one set and those in another, allowing for the comparison of sets with different cardinalities.

Single Drug Similarity (SDS)

To facilitate graph matching, a robust measure of similarity between individual drugs is required. The study compared two distinct approaches:

  1. $SDS_{2d}$: Traditional similarity based on 2D chemical structures (ECFP fingerprints).
  2. $SDS_{cm}$ (Co-medication Similarity): A novel approach based on clinical co-medication patterns. This method analyzes how frequently drugs are prescribed together in patients who did and did not experience a specific ADR. The hypothesis is that drugs used for similar therapeutic purposes or targeting similar biological pathways will exhibit similar co-medication "signatures."

Experimental Setup

The study utilized the FDA Adverse Event Reporting System (FAERS), a massive database of self-reported adverse events, focusing on myopathy (muscle disease) as the primary target ADR.

The researchers extracted over 64,000 "case" events (where myopathy was reported) and nearly 1.5 million "control" events. To validate the findings, the Side Effect Resource (SIDER) was used to identify individual drugs known to cause myopathy. A curated dataset ($D^*$) of 1,210 drugs and several thousand combinations was used for final model training and five-fold cross-validation.

Results and Discussion

The experimental results demonstrated that clinical wisdom can serve as a powerful proxy for biological similarity. The graph matching kernel ($K_{gm}$) combined with co-medication similarity ($SDS_{cm}$) achieved an AUC of 0.912, significantly outperforming traditional structural similarity and simpler baseline methods. The best-performing model reached an accuracy of 83.6%, compared to 76.5% for standard probabilistic kernels.

Key Insights

  • Beyond Pairs: The graph matching approach allows the model to compare a "cocktail" of three drugs to one of six drugs seamlessly, moving beyond the limitations of fixed-input models.
  • The Statin Signal: The model successfully identified that combinations involving statins (common cholesterol medications) were high-risk for myopathy, aligning with established medical literature. Crucially, it also identified risks in combinations where no single drug was a known offender.
  • Higher-Order Features: The researchers found that "order-2" kernels (which examine pairs within the larger set) performed better than "order-1" kernels, suggesting that the interaction between drug pairs is a critical building block for understanding larger combinations.

Conclusion

This research demonstrates that incorporating real-world co-medication patterns into a graph-based machine learning framework significantly improves the prediction of ADRs in complex drug combinations. While the study focused on myopathy, the methodology provides a robust foundation for predicting other adverse reactions. Future work may involve the use of Deep Learning to automatically learn drug representations from clinical data, potentially uncovering even more complex, non-linear drug relationships without the need for pre-defined kernels.

Summary updated

Abstract

Background: The problem of predicting whether a drug combination of arbitrary orders is likely to induce adverse drug reactions is considered in this manuscript. Methods: Novel kernels over drug combinations of arbitrary orders are developed within support vector machines for the prediction. Graph matching methods are used in the novel kernels to measure the similarities among drug combinations, in which drug co-medication patterns are leveraged to measure single drug similarities. Results: The experimental results on a real-world dataset demonstrated that the new kernels achieve an area under the curve (AUC) value 0.912 for the prediction problem. Conclusions: The new methods with drug co-medication based single drug similarities can accurately predict whether a drug combination is likely to induce adverse drug reactions of interest. Keywords: drug-drug interaction prediction; drug combination similarity; co-medication; graph matching
BibTeX
@article{chiang2020drug,
  author = {Wen-Hao Chiang and Li Shen and Lang Li and Xia Ning},
  title = {{Drug-drug interaction prediction based on co-medication patterns and graph matching}},
  journal = {Int. J. Comput. Biol. Drug Des.},
  year = {2020},
  url = {https://openreview.net/forum?id=1vIucAH5Ont},
}
Figure