基于功能网络信息传播预测疾病-miRNAs的关联

摘要 为了快速发现与疾病关联的miRNA， 基于功能网络信息传播，提出PMBP算法用于改进随机游走法，使用留一交叉验证评估了算法性能，最后进行案例分析.实验结果表明:对于尚未发现关联miRNA的疾病，随机游走法是失效的，而PMBP以疾病相似性作为先验信息，能够有效预测;对于已经关联miRNA的疾病，PMBP提高了预测性能，AUC值为0.866.对乳腺癌进行案例分析，预测的前50个miRNAs都被证实与乳腺癌相关，体现了PMBP算法的有效性.
Abstract：In order to quickly find out disease-related miRNAs， PMBP algorithm was proposed for improving random walk based on functional network information propagation. Leave-one-out cross validation was utilized to evaluate the performance of the algorithm and finally a case was analyzed. The results showed that random walk is ineffective for diseases that have not yet been associated with miRNAs， but the miRNA can be effectively predicted by using disease similarities as prior information. For the diseases known to be related with miRNAs， PMPB achieves a better performance and the corresponding AUC value is 0.866. In the case study of breast cancer， the predicted top 50 miRNAs are confirmed to be associated with breast cancer， which indicates the validity of PMBP.