# GraphMS: Drug Target Prediction Using Graph Representation Learning with Substructures

^{1}

^{2}

^{3}

^{4}

^{*}

## Abstract

**:**

## Featured Application

**Our work is to better discover potential DTI and provide new options for drug redirection.**

## Abstract

## 1. Introduction

- We apply the substructure embedding to DTI prediction, and remove certain noise in the graph network. The subgraph comparison strengthens the correlation between graph-level representation and subgraph representation to capture substructure information;
- We maximize the mutual information of node representation and graph–level representation. This allows the graph–level representation to contain more information about the node itself, and it will be more concentrated on the representative nodes in the embedded representation;
- Case study and comparison method experiments also show that our model is effective.

## 2. Related Work

#### 2.1. DTI Prediction

#### 2.2. Graph Representation Learning

## 3. Our Approach

#### 3.1. Problem Formulation

**Definition**

**1.**

Algorithm 1 GraphMS |

Input: A Heterogeneous Graph $G=({V}^{N},{E}^{R})$ |

Output: Drug–Protein Reconstruction Matrix M |

1: Perform feature integration on nodes whose node type $N\in \left(\right)open="\{"\; close="\}">drug,protein$ according to Equation (1); |

2: Select the relationship type $R\in \left(\right)open="\{"\; close="\}">drug-drug,protein-protein$ and convert to homogeneous graphs ${G}_{d}$,${G}_{p}$; |

3: while not Convergence do |

4: Shuffle X on row dimension to obtain $\tilde{X}$; |

5: for node type $N\in \left(\right)open="\{"\; close="\}">drug,protein$ do |

6: ${h}^{N}=Softmax\phantom{\rule{0.166667em}{0ex}}\left({A}^{N}{X}^{N}{W}_{1}^{N}\right)$; //Obtain Node-level Representation (Positive) |

7: ${\tilde{h}}^{N}=Softmax\phantom{\rule{0.166667em}{0ex}}\left({A}^{N}{\tilde{X}}^{N}{W}_{1}^{N}\right)$; //Obtain Node-level Representation (Negative) |

8: ${s}^{N}=\mathit{Pool}\phantom{\rule{0.166667em}{0ex}}\left(\frac{1}{n}{\sum}_{i=1}^{n}{h}_{i}^{N}\right)$; //Obtain Graph-level Representation |

9: Partition ${G}_{d}$,${G}_{p}$ nodes into k subgraphs ${G}_{{s}_{1}}^{N}$, ${G}_{{s}_{1}}^{N}$, ⋯, ${G}_{{s}_{k}}^{N}$ by METIS separately; |

10: for all each subgraph do |

11: Form the subgraph ${G}_{{s}_{k}}$ with nodes and edges into $\left(\right)$; |

12: Shuffle other nodes except nodes of the current subgraph and select k nodes randomly to obtain the corresponding negative subgraph $({\tilde{X}}_{{s}_{i}}^{N},{\tilde{A}}_{{s}_{i}}^{N})$; |

13: ${g}^{N}=Pool\phantom{\rule{0.166667em}{0ex}}\left(GC{N}_{sub}({X}_{s}^{N},{A}_{s}^{N})\right)$; //Obtain Substructure-level Representation (Positive) |

14: ${\tilde{g}}^{N}=Pool\phantom{\rule{0.166667em}{0ex}}\left(GC{N}_{sub}({\tilde{X}}_{s}^{N},{\tilde{A}}_{s}^{N})\right)$ //Obtain Substructure-level Representation (Negative); |

15: end for |

16: $Embeddin{g}^{N}=concat({h}^{N},{\left(concat({g}_{1}^{N},{g}_{2}^{N},\cdots ,{g}_{k}^{N})\right)}^{T})$; |

17: end for |

18: $U=Embeddin{g}^{drug},\phantom{\rule{1.em}{0ex}}V=Embeddin{g}^{protein}$; |

19: Compute the final loss and update parameters according to Equation (14); |

20: $M=U{W}_{3}\left(V{W}_{4}^{T}\right)$; |

21: end while |

22: return M |

#### 3.2. Information Fusion on Heterogeneous Graph

#### 3.3. GCN Encoder

#### 3.4. Mutual Information between Node–Level and Graph–Level Representation

#### 3.5. Mutual Information Between Graph–Level Representation and Substructure Representation

#### 3.6. Automatic Decoder for Prediction

## 4. Experiments and Results

#### 4.1. Datasets

#### 4.2. Experimental Settings

#### 4.3. Baselines

- NeoDTI [27] integrates the neighborhood information constructed by different data sources through a large number of information transmission and aggregation operations.
- DTINet [29] aggregates information on heterogeneous data sources, and can tolerate large amounts of noise and incompleteness by learning low–dimensional vector representations of drugs and proteins.
- LightGCN [30] simplified the design of GCN to make it more concise. This model only contains the most important part of GCN–neighborhood aggregation for collaborative filtering.
- GAT [26] proposes to use the attention mechanism to weight and sum the features of neighboring nodes. The weight of features of neighboring nodes depends entirely on the features of the nodes and is independent of the graph structure. GAT uses the attention mechanism to replace the fixed standardized operations in GCN. In essence, GAT just replaces the original GCN standardization function with a neighbor node feature aggregation function using attention weights.

#### 4.4. Comparative Experiment

#### 4.5. Ablation Experiment

#### 4.6. Case Study for Interpretability

## 5. Discussion

## 6. Patents

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Conflicts of Interest

## References

- Altae-Tran, H.; Ramsundar, B.; Pappu, A.S.; Pande, V. Low data drug discovery with one-shot learning. ACS Cent. Sci.
**2017**, 3, 283–293. [Google Scholar] [CrossRef] - Bleakley, K.; Yamanishi, Y. Supervised prediction of drug–target interactions using bipartite local models. Bioinformatics
**2009**, 25, 2397–2403. [Google Scholar] [CrossRef] [Green Version] - Ali, F.; El-Sappagh, S.; Islam, S.; Ali, A.; Attique, M.; Imran, M.; Kwak, K. An intelligent healthcare monitoring framework using wearable sensors and social networking data. Future Gener. Comput. Syst.
**2021**, 114, 23–43. [Google Scholar] [CrossRef] - Xie, Y.; Yao, C.; Gong, M.; Chen, C.; Qin, A. Graph convolutional networks with multi-level coarsening for graph classification. Knowl. Based Syst.
**2020**, 194, 105578. [Google Scholar] [CrossRef] - Ali, F.; El-Sappagh, S.; Islam, S.; Kwak, D.; Ali, A.; Imran, M.; Kwak, K. A smart healthcare monitoring system for heart disease prediction based on ensemble deep learning and feature fusion. Inf. Fusion
**2020**, 63, 208–222. [Google Scholar] [CrossRef] - Zhao, T.; Hu, Y.; Valsdottir, L.R.; Zang, T.; Peng, J. Identifying drug-target interactions based on graph convolutional network and deep neural network. Briefings Bioinform.
**2020**, 22, 2141–2150. [Google Scholar] [CrossRef] [PubMed] - Ashburner, M.; Ball, C.A.; Blake, J.A.; Botstein, D.; Butler, H.; Cherry, J.M.; Davis, A.P.; Dolinski, K.; Dwight, S.S.; Eppig, J.T.; et al. Gene ontology: Tool for the unification of biology. Nat. Genet.
**2000**, 25, 25–29. [Google Scholar] [CrossRef] [Green Version] - Kipf, T.N.; Welling, M. Semi-supervised classification with graph convolutional networks. arXiv
**2016**, arXiv:1609.02907. [Google Scholar] - Cao, S.; Lu, W.; Xu, Q. Grarep: Learning graph representations with global structural information. In Proceedings of the 24th ACM International on Conference on Information and Knowledge Management, Melbourne, VIC, Australia, 19–23 October 2015; pp. 891–900. [Google Scholar]
- Chiang, W.L.; Liu, X.; Si, S.; Li, Y.; Bengio, S.; Hsieh, C.J. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Anchorage, AK, USA, 4–8 August 2019; pp. 257–266. [Google Scholar]
- Jiao, Y.; Xiong, Y.; Zhang, J.; Zhang, Y.; Zhang, T.; Zhu, Y. Sub-graph Contrast for Scalable Self-Supervised Graph Representation Learning. arXiv
**2020**, arXiv:2009.10273. [Google Scholar] - Velickovic, P.; Fedus, W.; Hamilton, W.L.; Liò, P.; Bengio, Y.; Hjelm, R.D. Deep Graph Infomax. arXiv
**2019**, arXiv:1809.10341. [Google Scholar] - Park, C.; Han, J.; Yu, H. Deep multiplex graph infomax: Attentive multiplex network embedding using global information. Knowl. Based Syst.
**2020**, 197, 105861. [Google Scholar] [CrossRef] - Zhu, Y.; Che, C.; Jin, B.; Zhang, N.; Su, C.; Wang, F. Knowledge-driven drug repurposing using a comprehensive drug knowledge graph. Health Inform. J.
**2020**, 26, 2737–2750. [Google Scholar] [CrossRef] [PubMed] - Wan, F.; Zeng, J.M. Deep learning with feature embedding for compound-protein interaction prediction. bioRxiv
**2016**, 086033. [Google Scholar] [CrossRef] [Green Version] - Faulon, J.L.; Misra, M.; Martin, S.; Sale, K.; Sapra, R. Genome scale enzyme–metabolite and drug–target interaction predictions using the signature molecular descriptor. Bioinformatics
**2008**, 24, 225–233. [Google Scholar] [CrossRef] [Green Version] - Mei, J.P.; Kwoh, C.K.; Yang, P.; Li, X.L.; Zheng, J. Drug–target interaction prediction by learning from local information and neighbors. Bioinformatics
**2013**, 29, 238–245. [Google Scholar] [CrossRef] [PubMed] - Wen, M.; Zhang, Z.; Niu, S.; Sha, H.; Yang, R.; Yun, Y.; Lu, H. Deep-learning-based drug–target interaction prediction. J. Proteome Res.
**2017**, 16, 1401–1409. [Google Scholar] [CrossRef] - Hu, P.W.; Chan, K.C.; You, Z.H. Large-scale prediction of drug-target interactions from deep representations. In Proceedings of the 2016 International Joint Conference on Neural Networks (IJCNN), Vancouver, BC, Canada, 24–29 July 2016; pp. 1236–1243. [Google Scholar]
- Gao, K.Y.; Fokoue, A.; Luo, H.; Iyengar, A.; Dey, S.; Zhang, P. Interpretable Drug Target Prediction Using Deep Neural Representation. IJCAI
**2018**, 2018, 3371–3377. [Google Scholar] - Duvenaud, D.; Maclaurin, D.; Aguilera-Iparraguirre, J.; Gómez-Bombarelli, R.; Hirzel, T.; Aspuru-Guzik, A.; Adams, R.P. Convolutional networks on graphs for learning molecular fingerprints. arXiv
**2015**, arXiv:1509.09292. [Google Scholar] - Che, M.; Yao, K.; Che, C.; Cao, Z.; Kong, F. Knowledge-Graph-Based Drug Repositioning against COVID-19 by Graph Convolutional Network with Attention Mechanism. Future Internet
**2021**, 13, 13. [Google Scholar] [CrossRef] - Perozzi, B.; Al-Rfou, R.; Skiena, S. Deepwalk: Online learning of social representations. In Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, New York, NY, USA, 24–27 August 2014; pp. 701–710. [Google Scholar]
- Grover, A.; Leskovec, J. node2vec: Scalable feature learning for networks. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, CA, USA, 13–17 August 2016; pp. 855–864. [Google Scholar]
- Tang, J.; Qu, M.; Wang, M.; Zhang, M.; Yan, J.; Mei, Q. Line: Large-scale information network embedding. In Proceedings of the 24th International Conference on World Wide Web, Florence, Italy, 18–22 May 2015; pp. 1067–1077. [Google Scholar]
- Veličković, P.; Cucurull, G.; Casanova, A.; Romero, A.; Lio, P.; Bengio, Y. Graph attention networks. arXiv
**2017**, arXiv:1710.10903. [Google Scholar] - Wan, F.; Hong, L.; Xiao, A.; Jiang, T.; Zeng, J. NeoDTI: Neural integration of neighbor information from a heterogeneous network for discovering new drug–target interactions. Bioinformatics
**2019**, 35, 104–111. [Google Scholar] [CrossRef] [PubMed] - Karypis, G.; Kumar, V. A Fast and High Quality Multilevel Scheme for Partitioning Irregular Graphs. SIAM J. Sci. Comput.
**1998**, 20, 359–392. [Google Scholar] [CrossRef] - Luo, Y.; Zhao, X.; Zhou, J.; Yang, J.; Zhang, Y.; Kuang, W.; Peng, J.; Chen, L.; Zeng, J. A network integration approach for drug-target interaction prediction and computational drug repositioning from heterogeneous information. Nat. Commun.
**2017**, 8, 1–13. [Google Scholar] [CrossRef] [PubMed] [Green Version] - He, X.; Deng, K.; Wang, X.; Li, Y.; Zhang, Y.; Wang, M. Lightgcn: Simplifying and powering graph convolution network for recommendation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, Xi’an, China, 25–30 July 2020; pp. 639–648. [Google Scholar]
- Takeshima, H.; Niwa, T.; Yamashita, S.; Takamura-Enya, T.; Iida, N.; Wakabayashi, M.; Nanjo, S.; Abe, M.; Sugiyama, T.; Kim, Y.J.; et al. TET repression and increased DNMT activity synergistically induce aberrant DNA methylation. J. Clin. Investig.
**2020**, 130, 10. [Google Scholar] [CrossRef] [PubMed] - Rahman, M.M.; Tikhomirova, A.; Modak, J.K.; Hutton, M.L.; Supuran, C.T.; Roujeinikova, A. Antibacterial activity of ethoxzolamide against Helicobacter pylori strains SS1 and 26695. Gut Pathog.
**2020**, 12, 1–7. [Google Scholar] [CrossRef] [Green Version]

## Short Biography of Authors

Bo Jin Professor, PhD supervisor, innovative talents in colleges and universities in Liaoning Province, special consultant of Baidu Research Institute, outstanding member of CCF of China Computer Society, senior member of American ACM and IEEE. He graduated from Dalian University of Technology with a Ph.D., and visited Rutgers, the State University of New Jersey in the United States under the tutelage of Professor Xiong Hui, an authoritative scholar in the field of international big data. Served as Chair at ICDM, a top conference in the field of data mining for two consecutive years, and only two domestic experts hold relevant positions each year. The main research direction is deep learning, big data mining, artificial intelligence, and is committed to the analysis and mining methods of multi-source heterogeneous networked and serialized data. |

**Figure 1.**Framework overview. The model contains five parts: (1) Perform feature integration on heterogeneous graph. (2) GCN encoder is designed to obtain the node-level representation. (3) Feature-level shuffle is taken on the node representation to generate negative samples. Then maximize the mutual information between the node–level representation and graph–level representation. (4) Partition the homogeneous graphs into k subgraphs. Shuffle nodes except the current subgraph and select k nodes randomly to obtain the corresponding negative subgraph. Then maximize the mutual information between the substructure–level representation and graph–level representation. (5) A decoder takes latent vectors as input and output the reconstructed drug-protein matrix.

**Figure 3.**The process of retaining the sub-structure mutual information. Specifically, on the left is the process of obtaining the subgraph embedding. On the right is the process of generating negative subgraphs. We shuffle nodes except the current subgraph and select k nodes randomly to obtain the corresponding negative subgraph.

**Figure 5.**Visualizations of drug-target interaction network. We use Graph-M and Graph-S on the drug-target interaction dataset to learn drug/target embedding, which are visualized using t-SNE. Graph-M means the model uses only mutual information between node-level and graph-level representations. Graph-S means the model uses only mutual information between substructure and graph-level representations. The detail could be refered to the specific part of the above framework.

**Figure 6.**(

**a**) Diagram to compare ablation experiments under the condition in which all unknown drug–target interaction pairs were considered. (

**b**) Diagram to compare ablation experiments under the condition in which the ratio of positive to negative samples was 1:10.

**Figure 7.**Network visualization of the top 30 novel drug–target interactions predicted by GraphMS. Purple and blue nodes represent proteins and drugs, respectively. solid and dashed lines represent the known and predicted drug–target interactions.

Node Type | Drug | Protein | Disease |
---|---|---|---|

Number | 708 | 1512 | 5603 |

Edge Type | Number |
---|---|

Drug-Protein | 1923 |

Drug-Disease | 199,214 |

Protein-Disease | 1,596,745 |

Drug-Drug | 10,036 |

Protein-Protein | 7363 |

**Table 3.**Comparison of AUROC scores for various methods under different experimental settings. The best results in each column are in bold.

Model | Multi-View | Single-View | ||
---|---|---|---|---|

1:10 | 1:all | 1:10 | 1:all | |

GraphMS | 0.959 ± 0.002 | 0.943 ± 0.001 | 0.933 ± 0.003 | 0.914 ± 0.002 |

LightGCN | 0.940 ± 0.002 | 0.929 ± 0.001 | 0.922 ± 0.001 | 0.895 ± 0.002 |

GAT | 0.937 ± 0.001 | 0.927 ± 0.001 | 0.920 ± 0.001 | 0.893 ± 0.001 |

NeoDTI | 0.929 ± 0.003 | 0.919 ± 0.002 | 0.908 ± 0.001 | 0.880 ± 0.001 |

DTINet | 0.896 ± 0.001 | 0.862 ± 0.002 | 0.872 ± 0.001 | 0.867 ± 0.001 |

**Table 4.**Comparison of AUPR scores for various methods under different experimental settings. The best results in each column are in bold.

Model | Multi-View | Single-View | ||
---|---|---|---|---|

1:10 | 1:all | 1:10 | 1:all | |

GraphMS | 0.847 ± 0.002 | 0.622 ± 0.001 | 0.760 ± 0.002 | 0.594 ± 0.002 |

LightGCN | 0.834 ± 0.001 | 0.608 ± 0.001 | 0.734 ± 0.001 | 0.582 ± 0.001 |

GAT | 0.832 ± 0.002 | 0.608 ± 0.001 | 0.731 ± 0.001 | 0.581 ± 0.001 |

NeoDTI | 0.815 ± 0.003 | 0.587 ± 0.001 | 0.714 ± 0.001 | 0.559 ± 0.002 |

DTINet | 0.743 ± 0.001 | 0.452 ± 0.001 | 0.693 ± 0.002 | 0.313 ± 0.001 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Cheng, S.; Zhang, L.; Jin, B.; Zhang, Q.; Lu, X.; You, M.; Tian, X.
GraphMS: Drug Target Prediction Using Graph Representation Learning with Substructures. *Appl. Sci.* **2021**, *11*, 3239.
https://doi.org/10.3390/app11073239

**AMA Style**

Cheng S, Zhang L, Jin B, Zhang Q, Lu X, You M, Tian X.
GraphMS: Drug Target Prediction Using Graph Representation Learning with Substructures. *Applied Sciences*. 2021; 11(7):3239.
https://doi.org/10.3390/app11073239

**Chicago/Turabian Style**

Cheng, Shicheng, Liang Zhang, Bo Jin, Qiang Zhang, Xinjiang Lu, Mao You, and Xueqing Tian.
2021. "GraphMS: Drug Target Prediction Using Graph Representation Learning with Substructures" *Applied Sciences* 11, no. 7: 3239.
https://doi.org/10.3390/app11073239