TL;DR
This paper introduces LMSM, a novel computational framework for identifying lncRNA-miRNA sponge modules, which are groups of lncRNAs and mRNAs that collectively compete for miRNAs. LMSM leverages heterogeneous data to reveal complex regulatory mechanisms in breast cancer and outperforms existing methods.
What Problem Does This Paper Address?
Long non-coding RNAs (lncRNAs) can act as "miRNA sponges," sequestering microRNAs (miRNAs) and thereby regulating the expression of messenger RNAs (mRNAs). While existing methods focus on individual lncRNA-mRNA interactions, they often fail to capture the full picture of these complex regulatory networks. Specifically, the authors highlight that these methods don't fully account for how groups of lncRNAs and mRNAs might collectively compete for miRNAs, nor do they adequately consider the influence of shared miRNAs on these interactions. This gap limits our understanding of the modular regulatory mechanisms underlying diseases like breast cancer.
Methods
The core of the LMSM framework involves a two-stage process:
- Identifying lncRNA-mRNA co-expression modules: This stage uses the Weighted Gene Co-expression Network Analysis (WGCNA) method to group lncRNAs and mRNAs that exhibit similar expression patterns.
- Identifying lncRNA-miRNA sponge modules (LMSM modules): This stage filters the co-expression modules based on three key criteria derived from the authors' "miRNA sponge modular competition hypothesis":
- Significant sharing of miRNAs: The lncRNAs and mRNAs within a module must significantly share common target miRNAs. This is assessed using a hypergeometric test.
- High canonical correlation: The expression levels of the lncRNA group and the mRNA group within a module should be strongly and positively correlated. This is measured using canonical correlation analysis.
- Adequate sensitivity canonical correlation: This criterion assesses the influence of shared miRNAs on the competition between lncRNAs and mRNAs. It's calculated using a sensitivity correlation metric.
The framework also incorporates putative miRNA-target interaction databases to inform these criteria.
Datasets
The study utilizes the following datasets:
- The Cancer Genome Atlas (TCGA): Matched miRNA, lncRNA, and mRNA expression data, along with clinical data for breast cancer (BRCA).
- Putative miRNA-target interaction databases: Several high-confidence databases were used to retrieve miRNA-lncRNA and miRNA-mRNA interactions, including NPInter v3.0, DIANA-LncBase v2.0, miRTarBase v8.0, TarBase v7.0, and miRWalk v2.0.
- BRCA-related gene databases: DisGeNET, COSMIC, LncRNADisease v2.0, Lnc2Cancer v2.0, and MNDR v2.0 were used to identify BRCA-related genes.
- Experimentally validated lncRNA-related miRNA sponge interactions: A curated list of 581 interactions was collected for validation.
Key Insights and Ideas
- Modular Competition: The central hypothesis is that lncRNAs and mRNAs don't just compete individually, but rather in groups or modules, for shared miRNAs. This "tug-of-war" at a modular level is a novel perspective on gene regulation.
- Mediating miRNAs as Crosslinks: A significant finding is that most of the miRNAs involved in these sponge modules act as "crosslinks," connecting multiple LMSM modules. This suggests a highly interconnected regulatory network.
- BRCA Subtype Specificity: Many identified LMSM modules are not only implicated in breast cancer but are also specific to particular BRCA subtypes, highlighting the heterogeneity of the disease and the potential for subtype-specific therapeutic targets.
- LMSM as a Predictive Tool: The framework demonstrates utility beyond module identification, showing promise in predicting miRNA targets and offering a new avenue for investigating the biological functions and modular mechanisms of lncRNAs.
Key Results
- Identification of 17 LMSM modules in the BRCA dataset using WGCNA.
- Statistical Significance: All identified LMSM modules were found to be statistically significant.
- BRCA Implication: 10 out of 17 LMSM modules were functionally enriched in BRCA, and 15 were identified as module biomarkers.
- Subtype Specificity: A significant number of LMSM modules were found to be specific to different BRCA subtypes, with 41 unique subtype-specific modules identified.
- Classification Performance: LMSM modules significantly outperformed a baseline method in classifying BRCA subtypes, achieving higher Subset accuracy (0.7547 vs. 0.3800) and lower Hamming loss (0.0892 vs. 0.2480).
- Comparison with Graph Clustering: LMSM consistently outperformed a graph clustering-based strategy (GC) across various metrics, including % BRCA-related modules, % module biomarkers, classification performance, and validated interactions.
- Robustness: Using an alternative co-expression module identification method (SGFA) yielded similar results, indicating the robustness of the LMSM workflow.
- Prediction Power: LMSM predicted a large number of potential miRNA-target interactions, with a high percentage of novel miRNA-lncRNA interactions not identified by other methods.
Limitations and Future Work
- Experimental Validation: The authors acknowledge that experimentally validating entire LMSM modules is challenging due to their complexity. They suggest that biologists can select key lncRNAs and mRNAs within a module for targeted experimental validation.
- Generalizability: While the method is applied to breast cancer, the authors believe it can be extended to other human cancer datasets and other types of ncRNAs (e.g., circRNAs, pseudogenes).
- Parameter Tuning: LMSM is a parametric model, and its parameter settings can be adjusted based on specific research requirements, offering flexibility for future applications.
Summary updated