Liu Fuxing, Peng Lihong, Tian Geng, Yang Jialiang, Chen Hui, Hu Qi, Liu Xiaojun, Zhou Liqian
School of Computer Science, Hunan University of Technology, Zhuzhou, China.
Geneis (Beijing) Co. Ltd., Beijing, China.
Front Bioeng Biotechnol. 2020 Mar 17;8:131. doi: 10.3389/fbioe.2020.00131. eCollection 2020.
Recently, many studies have demonstrated that microRNAs (miRNAs) are new small molecule drug targets. Identifying small molecule-miRNA associations (SMiRs) plays an important role in finding new clues for various human disease therapy. Wet experiments can discover credible SMiR associations; however, this is a costly and time-consuming process. Computational models have therefore been developed to uncover possible SMiR associations. In this study, we designed a new SMiR association prediction model, RWNS. RWNS integrates various biological information, credible negative sample selections, and random walk on a triple-layer heterogeneous network into a unified framework. It includes three procedures: similarity computation, negative sample selection, and SMiR association prediction based on random walk on the constructed small molecule-disease-miRNA association network. To evaluate the performance of RWNS, we used leave-one-out cross-validation (LOOCV) and 5-fold cross validation to compare RWNS with two state-of-the-art SMiR association methods, namely, TLHNSMMA and SMiR-NBI. Experimental results showed that RWNS obtained an AUC value of 0.9829 under LOOCV and 0.9916 under 5-fold cross validation on the SM2miR1 dataset, and it obtained an AUC value of 0.8938 under LOOCV and 0.9899 under 5-fold cross validation on the SM2miR2 dataset. More importantly, RWNS successfully captured 9, 17, and 37 SMiR associations validated by experiments among the predicted top 10, 20, and 50 SMiR candidates with the highest scores, respectively. We inferred that enoxacin and decitabine are associated with mir-21 and mir-155, respectively. Therefore, RWNS can be a powerful tool for SMiR association prediction.
最近，许多研究表明，微小RNA（miRNA）是新的小分子药物靶点。识别小分子与miRNA的关联（SMiR）在为各种人类疾病治疗寻找新线索方面发挥着重要作用。湿实验可以发现可靠的SMiR关联；然而，这是一个成本高昂且耗时的过程。因此，已开发出计算模型来揭示可能的SMiR关联。在本研究中，我们设计了一种新的SMiR关联预测模型RWNS。RWNS将各种生物信息、可靠的负样本选择以及在三层异质网络上的随机游走集成到一个统一的框架中。它包括三个步骤：相似度计算、负样本选择以及基于在构建的小分子-疾病-miRNA关联网络上的随机游走进行SMiR关联预测。为了评估RWNS的性能，我们使用留一法交叉验证（LOOCV）和五折交叉验证，将RWNS与两种最先进的SMiR关联方法，即TLHNSMMA和SMiR-NBI进行比较。实验结果表明，在SM2miR1数据集上，RWNS在LOOCV下的AUC值为0.9829，在五折交叉验证下为0.9916；在SM2miR2数据集上，RWNS在LOOCV下的AUC值为0.8938，在五折交叉验证下为0.9899。更重要的是，RWNS在预测的得分最高的前10、20和50个SMiR候选物中，分别成功捕获了9个、17个和37个经实验验证的SMiR关联。我们推断，依诺沙星和地西他滨分别与mir-21和mir-155相关联。因此，RWNS可以成为SMiR关联预测的有力工具。
