Loading...

Heterogeneous Knowledge Graph-Based Drug Repurposing Using Graph Convolutional Networks

Gharizadeh Beyragh, Ali | 2022

129 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55707 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rabiee, Hamid Reza
  7. Abstract:
  8. Drug repurposing, which means finding a drug for a disease among approved drugs, is a solution for decreasing drug development time and cost. However,drug repurposing suggestions have been serendipitous so far. Therefore,proposing a systematic method for drug repurposing is crucial in terms of cost and human life savings. Recently, biological network-based methods for drug repurposing have had promising results among computational methods. But,these methods have some limitations too. First, the size and variety of data used in these methods are limited. Second, heterogeneous data is either not used or converted to homogeneous data leading to information loss in these methods. Finally, not using end-to-end methods and manual implementation of some parts of the methods have formed a need for an expert with domain knowledge. The last problem also makes it hard and sometimes impossible to change the input data. In this work, we propose a three-step method for knowledge graph-based drug repurposing. The first step is constructing a heterogeneous knowledge graph; The second step is graph node embedding with a heterogeneous graph transformer network; and the final step is computing relationship score with a fully connected network. This method does not have the stated limitations. Therefore, users will be able to manipulate input graphs and extract information from various entities and get their desired output using our method. Despite removing mentioned limitations, our method’s performance is comparable with previous methods. To evaluate the performance, we use the following metrics: the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPR). Our method showed a good performance using 5-fold crossvalidation technique (AUROC = 0.922, AUPR = 0.924). Additionally, we searched for experimental evidence for 10 best drug repurposing suggestions made by our method. We found experimental evidence for 8 suggestions; 1 suggestion can be true, and there is a need for more experimental investigation; for 1 suggestion, there was no evidence of relation, which can be a case study for experts in the field. We also demonstrated, with numerical and experimental validation, that our method can predict other types of relations, such as drug-protein relations and disease-protein relations. This feature makes our model a general link predictor in biological knowledge graphs.
  9. Keywords:
  10. Drug Repurposing ; Graph Convolutional Networks ; Knowledge Graph ; Heterogeneous Knowledge Graph ; Knowledge Graph-Based Methods

 Digital Object List

 Bookmark

...see more