Identification of miRNA-mRNA regulatory modules by exploring collective group relationships

Abstract

Background: microRNAs (miRNAs) play an essential role in the post-transcriptional gene regulation in plants and animals. They regulate a wide range of biological processes by targeting messenger RNAs (mRNAs). Evidence suggests that miRNAs and mRNAs interact collectively in gene regulatory networks. The collective relationships between groups of miRNAs and groups of mRNAs may be more readily interpreted than those between individual miRNAs and mRNAs, and thus are useful for gaining insight into gene regulation and cell functions. Several computational approaches have been developed to discover miRNA-mRNA regulatory modules (MMRMs) with a common aim to elucidate miRNA-mRNA regulatory relationships. However, most existing methods do not consider the collective relationships between a group of miRNAs and the group of targeted mRNAs in the process of discovering MMRMs. Our aim is to develop a framework to discover MMRMs and reveal miRNA-mRNA regulatory relationships from the heterogeneous expression data based on the collective relationships.

Results: We propose DIscovering COllective group RElationships (DICORE), an effective computational framework for revealing miRNA-mRNA regulatory relationships. We utilize the notation of collective group relationships to build the computational framework. The method computes the collaboration scores of the miRNAs and mRNAs on the basis of their interactions with mRNAs and miRNAs, respectively. Then it determines the groups of miRNAs and groups of mRNAs separately based on their respective collaboration scores. Next, it calculates the strength of the collective relationship between each pair of miRNA group and mRNA group using canonical correlation analysis, and the group pairs with significant canonical correlations are considered as the MMRMs. We applied this method to three gene expression datasets, and validated the computational discoveries.

Conclusions: Analysis of the results demonstrates that a large portion of the regulatory relationships discovered by DICORE is consistent with the experimentally confirmed databases. Furthermore, it is observed that the top mRNAs that are regulated by the miRNAs in the identified MMRMs are highly relevant to the biological conditions of the given datasets. It is also shown that the MMRMs identified by DICORE are more biologically significant and functionally enriched.

Fig. 1

DICORE workflow. Given the inputs of miRNA and mRNA expression profiles, we first derive an expression-based interaction weights matrix W using correlation test. We then compute two collaboration score matrices S and T from W for miRNAs and mRNAs based on their functional interaction similarities with common mRNAs (or miRNAs), respectively. Using these collaboration scores as input, we separately generate groups of miRNAs and groups of mRNAs at Stage 1 by an overlapping neighborhood expansion clustering algorithm, in which miRNAs or mRNAs are greedily added to (removed from) each cluster of miRNAs or mRNAs, respectively that maximize cohesiveness score of the cluster. Next in Stage 2, we apply canonical correlation analysis on the groups of miRNAs and groups of mRNAs to obtain significant collective group relationships, which are eventually the MMRMs with strength scores

Fig. 2

Confirmed interactions for miRNAs of the miR-200 family included in the top CORE ‘C1N65’ obtained from the NCI60. Red nodes are miRNAs, and green nodes are experimentally confirmed target mRNAs

Fig. 3

Predicted interactions for miRNAs included in the top CORE ‘C1N65’ obtained from the NCI60. Red nodes are miRNAs, yellow and white nodes are predicted (conserved) targets and poorly conserved targets of conserved miRNA families, respectively. Solid lines and dashed lines are used to represent links between miRNAs and their conserved targets and poorly conserved targets, respectively. The interactions are predicted by both DICORE and TargetScan