we also computed functional 1616113-45-1 similarity scores between XY1 miRNAs in the same cluster and between miRNAs not located in the same cluster using miRFunSim method. Statistical analyses showed that functional similarity scores among intracluster miRNA pairs, intercluster miRNA pairs and random miRNA pairs are also significantly different. The functional similarity scores of intracluster miRNA pairs are significantly higher than those of intercluster miRNA pairs and random miRNA pairs. Intercluster miRNA pairs also showed higher functional similarity scores than random miRNA pairs. These results suggested that our method is reliable and sensible to measure the functional relationship between miRNAs. To investigate the robustness of our miRFunSim method, we first performed above analysis for the predicted miRNA target and examine whether our method is still able to measure the associations between miRNAs. The resulting scores between miRNAs using predicted targets are shown in Fig. S1A. The functional similarity scores for intrafamily and intracluster miRNA pairs are significantly higher compared with interfamily and intercluster miRNA pairs. Then we further access our method by the removal of 5 and 10 network nodes in the protein interaction network randomly. As shown in Fig. S1B, C, the functional similarity scores of intrafamily and intracluster miRNA pairs are significantly higher than those of interfamily and intercluster miRNA pairs. To further evaluate the performance of our miRFunSim method for quantifying the associations between two miRNAs, we performed a validation analysis on experimentally verified miRNA-disease associations. It has been proven that miRNAs with similar functions tend to be involved in phenotypically similar disease, and miRNAs associated with common diseases are more related in function. Our validation analysis for performance of miRFunSim method was based on above notion. First, we obtained 270 high-quality experimentally verified miRNA-disease associations from Jiang��s study and 100 miRNAs whose target genes have been experimentally supported. For each disease, the functional similarity score between every two miRNAs associated with this disease were computed using the miRFunSim method as the testing case. For each testing case, 99 simulated miRNA pairs were generated and the target genes of simulated miRNA pairs were randomly sampled from all human protein-coding genes keeping the same size as the given testing case. The funct
