| 引用本文: | 汪颖,林家欣.基于混合协同过滤算法的随机游走[]模型预测lncRNA-疾病关联[J].生物信息学,2026,24(2):162-173. |
| WANG Ying,LIN Jiaxin.Prediction of lncRNA-disease associations by a random walk modelbased on a hybrid collaborative filtering algorithm[J].Chinese Journal of Bioinformatics,2026,24(2):162-173. |
|
| 摘要: |
| 近年来,研究显示lncRNA的突变和失调与许多复杂疾病密切相关。本文提出一种基于混合协同过滤算法的随机游走模型来预测lncRNA-疾病关联,简称HCRLDA。首先,分别将疾病语义相似性、lncRNA功能相似性及miRNA功能相似性分别与各自的高斯相似性集成,得到三个同构相似性网络。随后,结合已知的疾病-lncRNA、疾病-miRNA、lncRNA-miRNA关联信息构建了一个原始的多层异构相似性网络,由于已知的关联矩阵信息比较稀疏,其信息的完整性对模型的预测效果有着直接的影响,因此,引入了一种基于项目和用户的混合协同过滤推荐算法来更新原始异构网络。最后,在更新后的异构网络上实施重启随机游走算法来预测疾病关联的lncRNA,在十折交叉验证下HCRLDA的AUC平均值达到0.986 9,与以往的关联预测模型相比实验结果显示了该模型算法的有效性。 |
| 关键词: lncRNA-疾病 相似性 异构网络 协同过滤 随机游走模型 |
| DOI：10.12113/202501005 |
| 分类号:TP39 |
| 文献标识码:A |
| 基金项目: |
|
| Prediction of lncRNA-disease associations by a random walk modelbased on a hybrid collaborative filtering algorithm |
|
WANG Ying, LIN Jiaxin
|
|
(College of Science, Dalian Jiaotong University, Dalian 116028, Liaoning, China)[HJ1.5mm]
|
| Abstract: |
| In recent years, studies have shown that mutations and dysregulations of lncRNA are closely related to many complex diseases. In this paper, a random walk model based on a hybrid collaborative filtering algorithm is proposed to predict lncRNA-disease associations, referred to as HCRLDA. Firstly, disease semantic similarity, lncRNA functional similarity, and miRNA functional similarity are integrated with their respective Gaussian similarities to obtain three isomorphic similarity networks. Then, original multi-layer heterogeneous similarity networks are constructed by combining known disease-lncRNA, disease-miRNA, and lncRNA-miRNA association information. Due to the sparsity of the known association matrix information and the direct impact of information completeness on the prediction performance of the model. Therefore, a hybrid collaborative filtering recommendation algorithm based on items and users is introduced to update the original heterogeneous networks. Finally, the random walk with restart algorithm(RWR) is implemented on the updated heterogeneous network to predict diseases-related lncRNAs. The average AUC of HCRLDA reaches 0.9869 under ten-fold cross-validation, and experimental results demonstrate the effectiveness of the model compared to previous association prediction models. |
| Key words: lncRNA-disease Similarity Heterogeneous network Collaborative filtering Random walk model |
