Wang Wei, Lv Hehe, Zhao Yuan, Liu Dong, Wang Yongqing, Zhang Yu
Department of Computer Science and Technology, College of Computer and Information Engineering, Henan Normal University, Xinxiang, China.
Big Data Engineering Laboratory for Teaching Resources and Assessment of Education Quality, Xinxiang, China.
Front Bioeng Biotechnol. 2020 Apr 24;8:330. doi: 10.3389/fbioe.2020.00330. eCollection 2020.
The studies on drug-protein interactions (DPIs) had significant for drug repositioning, drug discovery, and clinical medicine. The biochemical experimentation () requires a long time and high cost to be confirmed because it is difficult to estimate. Therefore, a feasible solution is to predict DPIs efficiently with computers. We propose a link prediction method based on drug-protein interaction (DPI) local structural similarity (DLS) for predicting the DPIs. The DLS method combines link prediction and binary network structure to predict DPIs. The ten-fold cross-validation method was applied in the experiment. After comparing the predictive capability of DLS with the improved similarity-based network prediction method, the results of DLS on the test set are significantly better. Moreover, several candidate proteins were predicted for three approved drugs, namely captopril, desferrioxamine and losartan, and these predictions are further validated by the literature. In addition, the combination of the Common Neighborhood (CN) method and the DLS method provides a new idea for the integrated application of the link prediction method.
关于药物 - 蛋白质相互作用（DPI）的研究对药物重新定位、药物发现和临床医学具有重要意义。生化实验由于难以估计，需要很长时间和高昂成本才能得到证实。因此，一个可行的解决方案是利用计算机高效地预测DPI。我们提出了一种基于药物 - 蛋白质相互作用（DPI）局部结构相似性（DLS）的链接预测方法来预测DPI。DLS方法结合链接预测和二元网络结构来预测DPI。实验采用十折交叉验证法。在将DLS的预测能力与改进的基于相似性的网络预测方法进行比较后，DLS在测试集上的结果明显更好。此外，针对三种已获批药物卡托普利、去铁胺和氯沙坦预测了几种候选蛋白质，这些预测通过文献进一步得到验证。此外，共同邻居（CN）方法与DLS方法的结合为链接预测方法的综合应用提供了新思路。
