Graph Neural Network for Multirelational Link Prediction

Decagon is a graph convolutional neural network (GCN) approach for multirelational link prediction in multimodal graphs.

Decagon's graph convolutional neural network (GCN) model is a general approach for multirelational link prediction in any multimodal network. Decagon handles multimodal graphs with large numbers of edge types.

Here we specifically focus on using Decagon for problems in computational pharmacology. In particular, we model polypharmacy side effects. The use of multiple drugs, termed polypharmacy, is common to treat patients with complex diseases or co-existing medical 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. Discovering polypharmacy side effects is a challenge with significant implications for patient mortality and morbidity.

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 then predicts the exact side effect, if any, through which a given drug combination manifests clinically.

In computational pharmacology, Decagon creates an opportunity to use large molecular, pharmacological, and patient population data to flag and prioritize polypharmacy side effects for follow-up analysis via formal pharmacological studies.


Modeling polypharmacy side effects with graph convolutional networks.
Marinka Zitnik, Monica Agrawal and Jure Leskovec.
In Review, 2018. [arXiv] [biorXiv]

Multimodal graph of polypharmacy

Figure below shows an example graph of polypharmacy side effects derived from genomic and patient population data. A multimodal graph consists of protein-protein interactions, drug-protein targets, and drug-drug interactions encoded by 964 different polypharmacy side effects (i.e., edge types ri, i = 1, ..., 964). Side information is integrated into the model in the form of additional protein and drug feature vectors.

In the figure, highlighted graph neighbors of Ciprofloxacin (node C) indicate this drug targets four proteins and interacts with three other drugs. The graph encodes information that Ciprofloxacin (node C) taken together with Doxycycline (node D) or with Simvastatin (node S) increases the risk of bradycardia side effect (side effect type r2), and its combination with Mupirocin (M) increases the risk of gastrointestinal bleed side effect r1.

Decagon predicts associations between pairs of drugs and side effects (shown in red) with the goal of identifying polypharmacy side effects, i.e., side effects which cannot be attributed to either individual drug in the pair.

Decagon's graph convolutional neural network architecture

Decagon has two main components:

Figure below shows Decagon's encoder in the form of a GCN per-layer update for a single drug node (left), and Decagon's decoder in the form of a tensor factorization model (right).

Figure below show a batch of neural networks that compute embeddings of six drug nodes. In Decagon, neural networks differ from node to node but they all share the same set of relation type-specific trainable parameters.

Molecular network and polypharmacy data

File Description
bio-decagon-ppi.tar.gz Protein-protein interaction network
bio-decagon-targets.tar.gz Drug-target protein associations
bio-decagon-targets-all.tar.gz Drug-target protein associations culled from several curated databases
bio-decagon-combo.tar.gz Polypharmacy side effects in the form of (drug A, side effect type, drug B) triples
bio-decagon-mono.tar.gz Side effects of individual drugs in the form of (drug A, side effect type) tuples
bio-decagon-effectcategories.tar.gz Side effect categories


A TensorFlow implementation of Decagon is available on GitHub.