基于Laplacian正则化与双向随机游走的药物重定位方法

DRUG RELOCATION METHOD BASED ON LAPLACIAN NORMALIZATION AND TWO-PASS RANDOM WALK

传统的药物研制需要企业投入大量的人力物力和研发时间,而且要承担很高的失败风险。药物重定位可以为已有药物寻找新的适应症,极大地降低成本、提高研发效率,因此受到广泛重视。随着大规模生物表型数据的收集,基于计算方法的药物重定位表现出很大的潜力,其计算结果可以为生物学实验提供重要的方向性意见,所以研究优秀的药物重定位方法非常重要。提出LN-RWR算法,整合药物之间与疾病之间的相似度信息、药物和疾病的互作用信息构建异质网络,使用Laplacian正则化构造转移矩阵,在异质网路上进行随机游走,综合双向随机游走的结果进行药物重定位。实验结果表明,LN-RWR比现有方法效果更好,在不同数据集上都达到了较好的效果,而且案例分析结果表明LN-RWR的预测结果很多正在被生物学家们关注和研究。

Traditional drug discovery is time-consuming,costly and high risky. As a method of inferring new indications for the approved drugs,drug relocation is attracting more and more attention because successful repositioning can reduce cost and increase efficiency. With the information collection of biochemical and phenotypic data,computation-based drug relocation methods have shown great potential to predict drug-disease interactions accurately.The calculation results could provide guiding opinion for biological experiments,which represents the importance of studying drug relocation method. The LN-RWR algorithm is proposed to integrate disease-disease similarity and drugdisease associations into one heterogeneous network,then the Laplacian normalization is applied to build the transition matrix on the integrated network. We start drug-centric and disease-centric random walks with restart on the network respectively and integrate the scores in a comprehensive way to give the inferring results. Experiment results show that LN-RWR achieves better performance on different data sets,and the case study indicates that most predicted drugs are under investigation these years.