PDSM-LGCN: Prediction of drug sensitivity associated microRNAs via light graph convolution neural network
Lei Deng1, Ziyu Fan1, Hanlin Xu1
1School of Computer Science and Engineering, Central South University, Tianxin District, Hunan 410083, China.
Abstract:
Cancer has become one of the critical diseases threatening human life and health. The sensitivity difference of cancer drugs has always been a critical cause of the treatment come to nothing. Once drug resistance occurs, it will make the anticancer treatment or even various drugs ineffective. With the deepening of cancer research, a growing number of evidence shows that microRNA has a particular regulatory effect on the sensitivity of cancer drugs, which provides new research ideas. However, using traditional biological experiments to verify and discover the relations of microRNA-drug sensitivity is cumbersome and time-consuming, significantly slowing down cancer drug sensitivity's research progress. Therefore, this paper proposes a computational method (PDSM-LGCN) that spreads information through the high-order connection between cancer drug sensitivity and microRNA. At the same time, the model constructs an optimized-GCN as an embedding propagation layer to obtain the practical embeddings of microRNA and medicines. Finally, based on a collaborative filtering algorithm, the model brings the prediction score between microRNA and drug sensitivity. The results of fivefold cross-validation show that the AUC of PDSM-LGCN is 0.8872, and the AUPR is as high as 0.9026. At the same time, we also reproduced the five latest models of similar problems and compared the results. Our model has the best comprehensive effect among them. In addition, the reliability of PDSM-LGCN was further confirmed through the case study of Cisplatin and Doxorubicin, which can be used as a powerful tool for clinical and biological research. The source code and datasets can be obtained from https://github.com/19990915fzy/PDSM-LGCN/.
Insights
This study introduces PDSM-LGCN, a computational method to predict microRNA
Area of Science:
- Biomedical Informatics
- Computational Biology
- Genomics
Background:
- Drug resistance significantly hinders cancer treatment efficacy.
- MicroRNAs play a crucial role in regulating cancer drug sensitivity.
- Traditional experimental methods for microRNA-drug sensitivity discovery are slow and inefficient.
Purpose of the Study:
- To develop an efficient computational method for predicting microRNA-drug sensitivity.
- To accelerate the discovery of new cancer drug sensitivity regulators.
- To provide a tool for clinical and biological cancer research.
Main Methods:
- Proposed PDSM-LGCN, a computational model leveraging graph convolutional networks (GCN).
- Utilized high-order connections between microRNA and drug sensitivity for information propagation.
- Employed a collaborative filtering algorithm for prediction score generation.
Main Results:
- Achieved an Area Under the Curve (AUC) of 0.8872 and an Area Under the Precision-Recall Curve (AUPR) of 0.9026 in fivefold cross-validation.
- Outperformed five recent state-of-the-art models in comprehensive performance.
- Validated PDSM-LGCN's reliability through case studies with Cisplatin and Doxorubicin.
Conclusions:
- PDSM-LGCN is a powerful computational tool for predicting microRNA-drug sensitivity.
- The model significantly advances research progress in cancer drug sensitivity.
- PDSM-LGCN offers a reliable approach for clinical and biological applications in oncology.
