CoChat
Sign Up
Cited by 133Bioinformatics and Genomic NetworksComputational Drug Discovery MethodsAdvanced Graph Neural Networks

The DOI matches its Crossref record; 71 of the 71 references it cites were found in a registry or an index.

SkipGNN: predicting molecular interactions with skip-graph networks

Open access
Huang, Kexin, Xiao, Cao, Glass, Lucas M., Zitnik, Marinka, Sun, Jimeng·2020·Scientific Reports
DOI 10.1038/s41598-020-77766-9PMC PMC7713130
Open this paper in CoChat — free accountOpen PDF

Abstract

Molecular interaction networks are powerful resources for molecular discovery. They are increasingly used with machine learning methods to predict biologically meaningful interactions. While deep learning on graphs has dramatically advanced the prediction prowess, current graph neural network (GNN) methods are mainly optimized for prediction on the basis of direct similarity between interacting nodes. In biological networks, however, similarity between nodes that do not directly interact has proved incredibly useful in the last decade across a variety of interaction networks. Here, we present SkipGNN, a graph neural network approach for the prediction of molecular interactions. SkipGNN predicts molecular interactions by not only aggregating information from direct interactions but also from second-order interactions, which we call skip similarity. In contrast to existing GNNs, SkipGNN receives neural messages from two-hop neighbors as well as immediate neighbors in the interaction network and non-linearly transforms the messages to obtain useful information for prediction. To inject skip similarity into a GNN, we construct a modified version of the original network, called the skip graph. We then develop an iterative fusion scheme that optimizes a GNN using both the skip graph and the original graph. Experiments on four interaction networks, including drug–drug, drug–target, protein–protein, and gene–disease interactions, show that SkipGNN achieves superior and robust performance. Furthermore, we show that unlike popular GNNs, SkipGNN learns biologically meaningful embeddings and performs especially well on noisy, incomplete interaction networks.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41598-020-77766-9

Cites

71 of 71 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Direct versus skip similarity. (Left) Traditionally, an interaction between nodes A and B implies that A and B are similar and vice versa16. (Right) In contrast, in molecular interaction networks, directly interacting entities are not necessarily similar, which has been observed in numerous networks, including genetic interaction networks12,13 and protein–protein interaction networks14,15.
Figure 1 Direct versus skip similarity. (Left) Traditionally, an interaction between nodes A and B implies that A and B are similar and vice versa16. (Right) In contrast, in molecular interaction networks, directly interacting entities are not necessarily similar, which has been observed in numerous networks, including genetic interaction networks12,13 and protein–protein interaction networks14,15.CC BY · PubMed Central
Figure 2 Neural architecture of SkipGNN. (Left) SkipGNN constructs skip graph Gs (denoted by adjacency matrix As) based on the input graph G (denoted by adjacency matrix A) using Eq. (1). (Middle) Initial node embeddings, H⁽⁰⁾ and S⁽⁰⁾, are specified using side information (e.g., gene expression vectors if nodes represent genes) or generated using node2vec18. In SkipGNN, node embeddings are then propagated along edges of Gs and G and transformed through a series of computations (layers), which output powerful embeddings that can then be used for downstream prediction of interactions. Illustrated is a two-layer iterative fusion scheme. In the first layer, two GNNs with parameter weight matrices Wo⁽⁰⁾ and Ws⁽⁰⁾ (operating on A and As, respectively) are fused via weight matrices Wo^(^(′)(0)) and Ws^(^(′)(0)) based on Eq. (2). This completes computations in the first layer of SkipGNN, producing embeddings H⁽¹⁾ and S⁽¹⁾. In the second layer, those embeddings are transformed via Wo⁽¹⁾ and Ws⁽¹⁾ using Eq. (3), resulting in final embeddings E. (Right) Embeddings Ei and Ej of target nodes i and j are retrieved, concatenated, and then fed into a decoder (parameterized by Wd). Decoder returns pij, representing the probability that nodes i and j interact.
Figure 2 Neural architecture of SkipGNN. (Left) SkipGNN constructs skip graph Gs (denoted by adjacency matrix As) based on the input graph G (denoted by adjacency matrix A) using Eq. (1). (Middle) Initial node embeddings, H⁽⁰⁾ and S⁽⁰⁾, are specified using side information (e.g., gene expression vectors if nodes represent genes) or generated using node2vec18. In SkipGNN, node embeddings are then propagated along edges of Gs and G and transformed through a series of computations (layers), which output powerful embeddings that can then be used for downstream prediction of interactions. Illustrated is a two-layer iterative fusion scheme. In the first layer, two GNNs with parameter weight matrices Wo⁽⁰⁾ and Ws⁽⁰⁾ (operating on A and As, respectively) are fused via weight matrices Wo^(^(′)(0)) and Ws^(^(′)(0)) based on Eq. (2). This completes computations in the first layer of SkipGNN, producing embeddings H⁽¹⁾ and S⁽¹⁾. In the second layer, those embeddings are transformed via Wo⁽¹⁾ and Ws⁽¹⁾ using Eq. (3), resulting in final embeddings E. (Right) Embeddings Ei and Ej of target nodes i and j are retrieved, concatenated, and then fed into a decoder (parameterized by Wd). Decoder returns pij, representing the probability that nodes i and j interact.CC BY · PubMed Central
Figure 3 Predictive performance as a function of network incompleteness. SkipGNN provides robust result in varying fraction of missing edges. Fivefold average with 95% confidence interval for PR-AUC against various fractions of missing edges on four prediction tasks: drug–target interaction prediction (DTI), drug–drug interaction prediction (DDI), protein–protein interaction prediction (PPI) and gene–disease interaction prediction (GDI) on node2vec, Spectral Clustering (SC), Variational Graph Auto-Encoder (VGAE), Graph Convolutional Network (GCN), and SkipGNN. We omit DeepWalk as it has similar performance as node2vec. SkipGNN consistently shows the best performance even when networks are highly incomplete.
Figure 3 Predictive performance as a function of network incompleteness. SkipGNN provides robust result in varying fraction of missing edges. Fivefold average with 95% confidence interval for PR-AUC against various fractions of missing edges on four prediction tasks: drug–target interaction prediction (DTI), drug–drug interaction prediction (DDI), protein–protein interaction prediction (PPI) and gene–disease interaction prediction (GDI) on node2vec, Spectral Clustering (SC), Variational Graph Auto-Encoder (VGAE), Graph Convolutional Network (GCN), and SkipGNN. We omit DeepWalk as it has similar performance as node2vec. SkipGNN consistently shows the best performance even when networks are highly incomplete.CC BY · PubMed Central
Figure 4 Visualizations of drug–target interaction network. GCN does not distinguish drug and target gene as it only captures direct similarity whereas SkipGNN is able to distinct drug and target gene embeddings, confirming its ability to capture skip similarity. We use GCN and SkipGNN on the drug–target interaction dataset to learn drug/target embeddings, which are visualized using t-SNE.
Figure 4 Visualizations of drug–target interaction network. GCN does not distinguish drug and target gene as it only captures direct similarity whereas SkipGNN is able to distinct drug and target gene embeddings, confirming its ability to capture skip similarity. We use GCN and SkipGNN on the drug–target interaction dataset to learn drug/target embeddings, which are visualized using t-SNE.CC BY · PubMed Central
Figure 5 Visualizations of gene–disease interaction network. GCN does not distinguish disease and gene as it only captures direct similarity whereas SkipGNN is able to distinct disease and gene embeddings, confirming its ability to capture skip similarity. We use GCN and SkipGNN on the gene–disease interaction dataset to learn gene/disease embeddings, which are visualized using t-SNE.
Figure 5 Visualizations of gene–disease interaction network. GCN does not distinguish disease and gene as it only captures direct similarity whereas SkipGNN is able to distinct disease and gene embeddings, confirming its ability to capture skip similarity. We use GCN and SkipGNN on the gene–disease interaction dataset to learn gene/disease embeddings, which are visualized using t-SNE.CC BY · PubMed Central
Figure 6 The ability of skip graph and original graph to capture the network structure in the face of incomplete data. Skip graph can better preserve the network structure than the original graph, as evidenced by skip graph’s smaller relative error ("Robust learning on incomplete interaction network" section) than that of the original graph. This is true for all % of missing edges, indicating that skip graph can keep useful information about interaction structure even when networks are highly incomplete and many interactions are missing.
Figure 6 The ability of skip graph and original graph to capture the network structure in the face of incomplete data. Skip graph can better preserve the network structure than the original graph, as evidenced by skip graph’s smaller relative error ("Robust learning on incomplete interaction network" section) than that of the original graph. This is true for all % of missing edges, indicating that skip graph can keep useful information about interaction structure even when networks are highly incomplete and many interactions are missing.CC BY · PubMed Central
Figure 7 Network Visualization of four biomedical interaction networks.
Figure 7 Network Visualization of four biomedical interaction networks.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 65 references

Research works printed but not linked: 6

  1. Ribeiro, L. F., Saverese, P. H. & Figueiredo, D. R. struc2vec: Learning node representations from structural identity. In KDD 385–394 (2017).
  2. Xu, K., Hu, W., Leskovec, J. & Jegelka, S. How powerful are graph neural networks? In ICLR (2018).
  3. DuPont, P. Product Information. Coumadin (Warfarin). (DuPont Pharmaceuticals, Wilmington, 2000).
  4. Merck, C. I. Product Information. Belsomra (Suvorexant). (Merck & Company Inc., Whitehouse Station, 2014).
  5. Ligand, P. Product Information. Targretin (Bexarotene). (Ligand Pharmaceuticals, San Diego, 1999).
  6. Novartis, P. Product Information. Clozaril (Clozapine). (Novartis Pharmaceuticals, East Hanover, 1989).

Cited by

Sorted by citations

See all 133

Related

Metadata

Authors
Huang, K., Xiao, C., Glass, L. M., Zitnik, M., & Sun, J.
Affiliations
  • Harvard University (US)
  • IQVIA (United States) (US)
  • University of Illinois Urbana-Champaign (US)
Corresponding author
Sun, Jimeng
ORCID
Venue
Scientific Reports
Publisher
Springer Science and Business Media LLC
Published
2020-12-03
Volume
10
Issue
1
Pages
21092-21092
ISSN
2045-2322 (ISSN-L)
DOI
10.1038/s41598-020-77766-9
PMC
PMC7713130
OpenAlex
W3108458441
Cited by
133
References
65
Citations by year
2026: 14 · 2025: 28 · 2024: 34 · 2023: 18 · 2022: 24 · 2021: 12 · 2020: 3
FWCI
4.62
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • arXiv (Cornell University) (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (cc-by)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Physical Sciences › Computer Science › Artificial Intelligence › Advanced Graph Neural Networks
Keywords
graph neural networks, molecular interaction prediction, second-order interactions, skip graphs, drug-drug interactions, drug-target interactions, protein-protein interaction, gene-disease interaction, node embeddings

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review