As further corroborating test, we observed that, when search ing the target coding genes of homologous miRNAs the list of predicted targets is identical for all miRNAs. Moreover, we notice that only two homologous necessary groups of miRNAs in the cluster are not part of F3. If we look at their sequence in detail we observe that they are very similar to miR 20a with only two mismatches, one in the loop and one after the supplemen tary pairing region. This can represent a partial functional redundancy since all the known key regions in target recognition are identical. Conversely, miR 92 does not share any significant homology with the other members of the cluster. Taking into consideration all the redundancies in the clusters, most of the transcript targets in F3 are probably under the regulation effect of the expressed miR NAs.

It is worth noting that a cross hybridization effect in miRNAs could be considered the mechanism responsible for these association in clusters. But, as reported by the authors of the dataset, each primer and probe con tained zip coded sequences specifically assigned to each miRNA to increase the specificity of each reaction so that even small differences in miRNA were amplified and detected. So, this artifact can be discarded as explanation for the emerging of clusters of miRNA. Statistical Rele vance, Interestingly, in F3, only 2 miRNAs out of 7 do not belong to any of these two clusters. Their role was shown respectively to be related to the molecular pathogenesis of ovarian cancer as well as to schizophrenia and Human T cell leuke mia Virus 1 transformation.

Six more miRNAs that belong to these two clus ters could not be part of our analysis, as they were not part of Lius original dataset. Given the high density of miRNAs in these clusters, we used the hypergeometric dis tribution to compute the probability associated with the hypothesis that a random sampling would give the same result in terms of number of cluster members in cluster miR 17 92, in cluster miR 106 363 and in both. The reference group for computing the probability consists of the total number of detected miR NAs. The resultant probabilities were Bonferroni cor rected and were equal to 3. 6 �� 10?3, 0. 045 and 2. 3 �� 10?7 respectively. All three are statistically significant.

Speculations on Molecular Clinical Implications Ultimately, we speculated on how the two clusters that emerge in F3 can, along with the molecular analysis performed on F1, discriminate between gliosarcomas and non gliosarcomas. This choice is due to the fact that our analysis has shown that the combination of fac tors that carry the more coherent functional information was GSK-3 the com bination able to discriminate glioscarcomas from other tumors. Believing that such a coherence could hide strong biological meanings we focused on gliosarcomas the efforts to detect emergent properties.