To verify the results from the hypergeometric tests on the experimental data, we generated 1000 randomized samples of non-modulated miRNAs and 1000 randomized samples of non-modulated mRNAs. The size of the samples were fixed and determined by the number of corresponding modulated miRNAs and mRNAs in the experimental dataset. These randomized samples were paired into a 1000 sets of randomized miRNAs and mRNAs and iteratively subjected to the same statistical analysis workflow as the experimental dataset. These represent a set of 1000 randomized trials, which represent our randomized control group. Proportions of interest (proportion of putative targets among randomly selected mRNAs; proportion of randomly selected mRNAs being targeted by more than 1 randomly selected miRNAs; proportion of randomly selected miRNAs that have an over-represented number of randomly selected mRNA targets; and proportion of randomly selected mRNAs that have an over-represented numbers of randomly selected miRNAs targeting them) were recorded for each trial, summarized after completion of 1000 trials, and statistically compared to corresponding experimental proportions using the z-score test (z-test), also implemented in the R environment. The download of miRecords’ predicted target information for every human miRNA accessible through the website was done using the RCurl package implemented in R.
