A computational drug repositioning model based on hybrid similarity side information powered graph neural network

https://doi.org/10.1016/j.future.2021.06.018Get rights and content

Highlights

  • We use GNN to capture the hidden feature representations of drugs and diseases.

  • We use dimensionality reduction algorithms to overcome the cold-start problem.

  • Experimental results show the superiority on two real drug-disease datasets.

Abstract

Computational drug repositioning technology aims to rediscover the potential use of drugs already on the market and can significantly accelerate the traditional drug development process, reducing significant drug development costs and drug development instability

In this work, in order to capture valid and robust hidden feature representations of drugs and diseases, we introduce a new computational drug relocation model, HSSIGNN, based on hybrid similarity side information powered graph neural network, by drawing on the application of graph neural networks and Side information in recommender systems. Its advantage is to utilize the learning capability of graph neural networks to capture the effective hidden feature representation of drugs and diseases, which is used to infer the probability of whether a drug can treat the disease of interest, as a way to improve the generalization capability of the model. In addition, dimensionality reduction algorithms and side information of drugs and diseases are used to overcome the cold start problem encountered by traditional computational drug relocation models. Finally, the experimental results of the proposed model on two real drug–disease association datasets are analyzed to verify its superiority and effectiveness. Comprehensive experimentations on several real-world datasets show the efficiency of HSSIGNN.

Introduction

Medication exploration is really a huge expense as well as huge threat procedure [1], [2]. Among one of the most essential action in medication project is composed in presuming possible signs for unique particles as well as in the repositioning of authorized medications [3], [4]. Particularly medication repurposing has the advantage of beginning with well-characterized particles, thus decreasing the dangers in medical stages as well as the expense of tests [5], [6].

Computational drug repositioning technology aims to rediscover the potential use of drugs already in the market, and it can significantly accelerate the traditional drug development process, reducing significant drug development costs and drug development instability [7], [8]. Computational drug repositioning techniques have attracted the attention of a large number of researchers and companies due to their intrinsic and significant economic value [9].

In this work, in order to be able to obtain effective and robust hidden feature representations of drugs and diseases, we introduced a new computational drug repositioning model, HSSIGNN, based on hybrid similarity side information powered graph neural network, drawing on the application of graph neural networks [10] and side information [11] in recommender systems. Firstly, in order to obtain the effective hidden features of drugs and diseases, the HSSIGNN model draws on the graph neural network operator in High-Order GNN , which is used to compute the effective hidden feature values of drugs and diseases. Secondly, to be able to obtain a robust hidden feature representation, the HSSIGNN model uses the drug–disease side information to extract another hidden feature representation by feding the drug-to-disease similarity matrix and the disease-to-disease similarity matrix into the PCA algorithm. Then, in order to be able to consider the contribution of both hidden feature representations to the final predicted values, the HSSIGNN model performs a splicing operation of the two hidden features of the drug and the disease, and subsequently inputs the spliced hidden features of the drug and the disease into a three-layer autoencoder to extract the respective final hidden feature representations. Finally, the final hidden feature representation of the drug and the disease is element-wise multiplied and fed into a single layer fully connected network to obtain the final predicted value, the magnitude of which represents the probability of the drug being able to treat the disease.

The main contributions made by this work are as follows.

  • We use the learning capability of graph neural networks to capture the valid hidden feature representations of drugs and diseases, which are used to infer the probability of whether a drug can treat the disease of interest, as a way to improve the generalization capability of the model.

  • We use dimensionality reduction algorithms and drug or disease side information to overcome the cold-start problem experienced by traditional computational drug repositioning models.

  • We verified the superiority and validity of the model proposed in this work by analyzing its experimental results on two real drug–disease association datasets.

The subsequent sections of this work are structured as follows. In Section 2, “Related Work ”, we present the results of the current mainstream computational drug repositioning models. Then, in Section 3, “Method”, we will analyze the implementation details of the HSSIGNN model. In Section 4, “Experiment and Discussion”, we will discuss the experimental results of the proposed HSSIGNN model on several real-world drug–disease association datasets and compare the results with other classical classification models. Finally, in Chapter Section 5, “Conclusion”, we will conclude our work.

Section snippets

Related work

Over the last few years, research workers have actually proposed a range of computational medication repurposing methods [11], [12], [13], [14], [15], [16], [17], [18], [19], which includes graph-based approaches, matrix factorization based techniques, Collective filtering system and so on.

Based upon the presumption that resembling medications are generally connected with resembling illness as well as the other way around, Luo et al. [12] introduced an unique computational approach called

Method

In this section, we will analyze the implementation details of each part of the HSSIGNN model and the related formulas. Fig. 1 shows the algorithm flowchart of the HSSIGNN model, which contains four input matrices, namely, the drug–disease association matrix R and its transpose matrix RT, the drug-drug similarity matrix DrugSim, and the disease-disease similarity matrix SicknessSim.

The HSSIGNN model first inputs R and RT into the Truncated SVD model, and the output values are used as the

Results and discussion

In this section, we will explore the experimental results of the HSSIGNN model proposed in this work on several real-world drug–disease association datasets, including the analysis of the results on some important parameters and the comparison results with other classical classification models. Firstly, the relevant datasets used for the experiments will be presented in Section 4.1. Secondly, in Section 4.2, the evaluation metrics used in the experiments are presented. Then in Section 4.3, the

Conclusion

Computational drug repositioning technology aims to rediscover the potential use of drugs already on the market and can significantly accelerate the traditional drug development process, reducing significant drug development costs and drug development instability

In this work, in order to capture valid and robust hidden feature representations of drugs and diseases, we introduce a new computational drug relocation model, HSSIGNN, based on hybrid similarity side information powered graph neural

CRediT authorship contribution statement

Sumin Li: Completed the algorithm framework, Writing - original draft. Xiuqin Pan: Proposed the idea, Checked the language and writing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Sumin Li received the M.S. degree in 2003. She is currently a Lecturer at the School of Information Engineering, Minzu University of China, Beijing, China. His current research interests include big data, intelligent computing and intelligent information processing and system.

References (25)

  • MartinezV. et al.

    Drugnet: network-based drug–disease prioritization by integrating heterogeneous data

    Artif. Intell. Med.

    (2015)
  • ShimJ.S. et al.

    Recent advances in drug repositioning for the discovery of new anticancer drugs

    Int. J. Biol. Sci.

    (2014)
  • DicksonM. et al.

    Key factors in the rising cost of new drug discovery and development

    Nat. Rev. Drug Discov.

    (2004)
  • TamimiN.A. et al.

    Drug development: from concept to marketing!

    Nephron Clin. Pract.

    (2009)
  • PushpakomS. et al.

    Drug repurposing: progress, challenges and recommendations

    Nat. Rev. Drug Discov.

    (2019)
  • AshburnT.T. et al.

    Drug repositioning: identifying and developing new uses for existing drugs

    Nat. Rev. Drug Discov.

    (2004)
  • NosengoN.

    Can you teach old drugs new tricks?

    Nature

    (2016)
  • PritchardJ.-L.E. et al.

    Enhancing the promise of drug repositioning through genetics

    Front. Pharmacol.

    (2017)
  • YellaJ.K. et al.

    Changing trends in computational drug repositioning

    Pharmaceuticals

    (2018)
  • LuoH. et al.

    Biomedical data and computational models for drug repositioning: A comprehensive review

    Brief. Bioinform.

    (2019)
  • C. Morris, M. Ritzert, M. Fey, W.L. Hamilton, J.E. Lenssen, G. Rattan, M. Grohe, Weisfeiler and leman go neural:...
  • YangX. et al.

    Additional neural matrix factorization model for computational drug repositioning

    BMC Bioinformatics

    (2019)
  • Cited by (9)

    • Machine learning methods in drug design

      2023, Cheminformatics, QSAR and Machine Learning Applications for Novel Drug Development
    • Flowlet-level multipath routing based on graph neural network in OpenFlow-based SDN

      2022, Future Generation Computer Systems
      Citation Excerpt :

      The purpose of GNN is to aggregate part or all of the graph structure information at nodes through loop iteration, so as to realize the end-to-end processing of complex graph data by the deep learning model. At present, GNN structures represented by Graph Convolutional Neural Network, Graph Attention Network, Graph Autoencoder and Spatial–Temporal Graph Neural Network have been applied in computer vision, recommendation system, chemical analysis and other scenarios and achieved surprising results [20,21]. Inspired by the existing works, in this paper we use GNN to achieve the prediction of network link delay.

    • Self-Supervised Learning for Label Sparsity in Computational Drug Repositioning

      2023, IEEE/ACM Transactions on Computational Biology and Bioinformatics
    View all citing articles on Scopus

    Sumin Li received the M.S. degree in 2003. She is currently a Lecturer at the School of Information Engineering, Minzu University of China, Beijing, China. His current research interests include big data, intelligent computing and intelligent information processing and system.

    Xiuqin Pan received the B.E. degree in electric technology and the M.E. degree in power system and automation specialty from Zhengzhou University, Zhengzhou, China, in 1994 and 1999, respectively, and the Ph.D. degree in control theory and control engineering from the Beijing Institute of Technology, in 2002. She is currently a Professor with the School of Information Engineering, Minzu University of China. Her current research interests focus on parallel algorithm and intelligent systems.

    View full text