Figure 2.
The combinatorial metaMIR analysis extracts miRNA candidates above random levels. Genes were selected randomly from the entire pool of genes in the miRNA–target reference set (Ran) or from the experimentally validated data (Pos), where, in the latter case, it is expected that a miRNA should co-regulate the random selection of genes. (A) A difference is observed in the count of miRNA–gene-cluster sets returned in the Pos versus the Ran analyses. At smaller cluster sizes, the counts are similar, but with increasing set sizes, many fewer co-regulated clusters are found in the random data. (B) Cumulative distribution plots of the final adjusted standardised scores (Sadj, Equation 4) illustrate the increase in performance above random gene selections, with higher scores returned for the positive, over the random analysis. (C) The analysis was repeated using mixed lists, where only a percentage of the gene list (Pos Seed) was derived from the Pos set and the rest were random selections. Even at 33%, scores can be distinguished from those of the purely random and the maximum difference occurs at a score threshold of 1.1. (D) MetaMIR is blinded to the miRNA used in the generation of the random sets. In each analysis, the source miRNA was recovered in the results; its mean rank (AvRank) decreased slightly as the proportion of the seed decreased. Capture of this miRNA among the top 10 candidates returned is poor at lower seeding, as its score may be overtaken by clusters of a size similar to the seed (five genes) found at random.