top of page

AI Case Study

Researchers at Stanford University develop an approach for modeling polypharmacy side effects with graph convolutional networks outperforming baselines by up to 69%

Researchers at the University of Stanford have developed a model to predict the side effects of using drug combinations, termed polypharmacy. Side effects are formulated as a multirelational link prediction problem in a two-layer multimodal network consisting of drugs and proteins. Decagon, the convolutional graph neural network model, is able to accurately predict the exact side effects of polypharmacy outperforming baselines by up to 69%.

Industry

Healthcare

Pharmaceuticals And Biotech

Project Overview

"The approach constructs a multimodal graph of protein–protein interactions, drug–protein target interactions and the polypharmacy side effects, which are represented as drug–drug interactions, where each side effect is an edge of a different type. Decagon is developed specifically to handle such multimodal graphs with a large number of edge types. Our approach develops a new graph convolutional neural network for multirelational link prediction in multimodal networks.

To motivate our model, we first perform exploratory analysis leading to two important observations (Section 3). First, we find that co-prescribed drugs (i.e. drug combinations) tend to have more target proteins in common than random drug pairs, suggesting that drug-target protein information contains valuable information for drug combination modeling. Second, we find that it is important to consider a map of protein–protein interactions in order to be able to model characteristics of drugs with common side effects. These observations motivate the development of Decagon to make predictions about which drug pairs will interact and what will the exact type of the interaction/side effect be (Section 4).
Decagon develops a new graph auto-encoder approach (Hamilton et al., 2017a), which allows us to develop an end-to-end trainable model for link prediction on a multimodal graph. In contrast, previous graph-based approaches for link prediction tasks in biology (e.g. Chen et al. 2016b; Huang et al. 2014b; Zong et al. 2017) employ a two-stage pipeline, typically consisting of a graph feature extraction model and a link prediction model, both of which are trained separately. Furthermore, the crucial distinguishing char- acteristic of Decagon is the multirelational link prediction ability allowing us to capture the interdependence of different edge (side ef- fect) types, and to identify which out of all possible edge types exist between any two drug nodes in the graph."

Reported Results

"Unlike approaches limited to predicting simple drug–drug interaction values, Decagon can predict the exact side effect, if any, through which a given drug combination manifests clinically. Decagon ac- curately predicts polypharmacy side effects, outperforming baselines by up to 69%."

Technology

The "model is a convolutional graph neural network that operates in a multirelational setting."

Function

R And D

Core Research And Development

Background

"The use of drug combinations, termed polypharmacy, is common to treat patients with complex diseases or co-existing conditions. However, a major consequence of polypharmacy is a much higher risk of adverse side effects for the patient. Polypharmacy side effects emerge because of drug–drug interactions, in which activity of one drug may change, favorably or unfavorably, if taken with another drug. The knowledge of drug interactions is often limited because these complex relationships are rare, and are usually not observed in relatively small clinical testing. Discovering polypharmacy side effects thus remains an important challenge with significant implications for patient mortality and morbidity."

Benefits

Data

"We formulate the polypharmacy side effect identification problem as a multirelational link prediction problem in a two-layer multimodal graph/network of two node types: drugs and proteins. We construct two-layer multimodal network as follows (Fig. 1). Protein–protein interaction network describes relationships between proteins. Drug–drug interaction network contains 964 different types of edges (one for each side effect type) and describes which drug pairs lead to which side effects. Lastly, drug-protein links describe the proteins targeted by a given drug."

bottom of page