DRLM: a robust drug representation learning method and its applications

1 Introduction

High-quality and robust drug representations can broaden the understanding of pharmacology, and improve the modeling of multiple drug-related prediction tasks, which further facilitates drug development. DRLM is such a framework to learn drug representations with integrating gene expression profiles of drug-related cells and the therapeutic use information of drugs. And the learned drug representations by DRLM can be used for various drug-related downstream tasks, such as drug-disease association prediction, drug-drug interaction prediction and miRNA-drug resistance association prediction tasks.

2 Overview

Here we provide an implementation of DRLM with Python. The repository is organised as follows:

• dataset/drugMiRNAData/ contains the necessary dataset files for predicting miRNA-drug resistance associations, which comes from the paper of Niu et al.¹ The drug-disease association dataset and the drug-drug interaction dataset can be obtained from the paper of Zhang et al.² and the paper of Liu et al.³, respectively.
• ourMethod/method/ contains the code scripts of our proposed method. Details are as follows:
  • representationLearning.py consists of three modules, namely, the stacked autoencoder, the iterative clustering module and the therapeutic use discriminator.
  • expCV.py performs 5-fold cross-validation experiments for downstream tasks.
  • experiments.py puts all code scripts together and executes a full training run.

Below is an example command line for evaluating the model on the miRNA-drug resistance association prediction task:

Step 1 (to learn the drug representations):

python experiments.py --expName representationLearning

Step 2 (to evaluate on the miRNA-drug resistance association prediction task):

python experiments.py --expName expCV --dataName drugMiRNA --clfName RF

We used the DESC algorithm during constructing our framework. We thank them a lot for their code sharing!

3 Requirements

python==3.7.1

keras==2.2.5

tensorflow==1.15.0

anndata==0.7.8

scanpy==1.7.2

sklearn==0.24.2

numpy==1.21.2

pandas==1.2.0

4 Citation

@INPROCEEDINGS{Zhao2021robust,

author={Zhao, Cecheng and Huang, Ziyang and Wang, Hui and Fu, Haitao and Wang, Dong and Gao, Yingjie and Zhu, Haotian and Niu, Xiaohui and Zhang, Wen},

booktitle={2021 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)},

title={A robust drug representation learning model for eliminating cell specificity in gene expression profile and its application},

year={2021},

pages={1191-1196},

doi={10.1109/BIBM52615.2021.9669385}

}

5 Contact

• Please feel free to contact us if you need any help: [email protected] OR [email protected]
• Attention: Only real name emails will be replied. Please provide as much detail as possible about the problem you are experiencing.
• 注意：只回复实名电子邮件。请尽可能详细地描述您遇到的问题，可以附上截图等。

[1] Niu, Y., Song, C., Gong, Y., & Zhang, W. (2022). MiRNA-Drug Resistance Association Prediction Through the Attentive Multimodal Graph Convolutional Network. Frontiers in pharmacology, 12, 799108. https://doi.org/10.3389/fphar.2021.799108

[2] Zhang, W., Yue, X., Lin, W. et al. (2018). Predicting drug-disease associations by using similarity constrained matrix factorization. BMC Bioinformatics 19, 233. https://doi.org/10.1186/s12859-018-2220-4

[3] Liu, S., Zhang, Y., Cui, Y., Qiu, Y., Deng, Y., Zhang, Z. M., & Zhang, W. (2022). Enhancing drug-drug interaction prediction using deep attention neural networks. IEEE/ACM Transactions on Computational Biology and Bioinformatics.