Limits...
Towards computational prediction of microRNA function and activity.

Ulitsky I, Laurent LC, Shamir R - Nucleic Acids Res. (2010)

Bottom Line: Our analysis is based on a novel compendium of experimentally verified miRNA-pathway and miRNA-process associations that we constructed, which can be a useful resource by itself.Our method also predicts novel miRNA-regulated pathways, refines the annotation of miRNAs for which only crude functions are known, and assigns differential functions to miRNAs with closely related sequences.Applying our approach to groups of co-expressed genes allows us to identify miRNAs and genomic miRNA clusters with functional importance in specific stages of early human development.

View Article: PubMed Central - PubMed

Affiliation: Blavatnik School of Computer Science, Tel Aviv University, Tel Aviv, Israel. ulitsky@wi.mit.edu

ABSTRACT
While it has been established that microRNAs (miRNAs) play key roles throughout development and are dysregulated in many human pathologies, the specific processes and pathways regulated by individual miRNAs are mostly unknown. Here, we use computational target predictions in order to automatically infer the processes affected by human miRNAs. Our approach improves upon standard statistical tools by addressing specific characteristics of miRNA regulation. Our analysis is based on a novel compendium of experimentally verified miRNA-pathway and miRNA-process associations that we constructed, which can be a useful resource by itself. Our method also predicts novel miRNA-regulated pathways, refines the annotation of miRNAs for which only crude functions are known, and assigns differential functions to miRNAs with closely related sequences. Applying our approach to groups of co-expressed genes allows us to identify miRNAs and genomic miRNA clusters with functional importance in specific stages of early human development. A full list of the predicted mRNA functions is available at http://acgt.cs.tau.ac.il/fame/.

Show MeSH

Related in: MedlinePlus

Performance of methods for enrichment detection on co-expression clusters. Out of the 1323 possible miRNA–cluster pairs, those with a correlation of r > 0.5 or r < −0.5 between the miRNA and the average mRNA expression were marked as ‘high’ (∼10% for each direction). The plots show the fraction of the 100 most significant miRNA–cluster pairs found by FAME and the HG test that fell into the ‘high’ category.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2926627&req=5

Figure 3: Performance of methods for enrichment detection on co-expression clusters. Out of the 1323 possible miRNA–cluster pairs, those with a correlation of r > 0.5 or r < −0.5 between the miRNA and the average mRNA expression were marked as ‘high’ (∼10% for each direction). The plots show the fraction of the 100 most significant miRNA–cluster pairs found by FAME and the HG test that fell into the ‘high’ category.

Mentions: According to the currently accepted model, miRNAs function as repressors of gene expression, and thus their expression patterns are expected to be anti-correlated with those of their targets (50,51) and correlated with their anti-targets [genes depleted of miRNA target sites (18)]. However, it is often difficult to identify such effects in matched miRNA/mRNA expression datasets. For example, in the SCD, the average Pearson correlation between expression of miRNAs and their TargetScan targets is 0.009. The uneven 3′ UTR lengths of genes highly expressed in different stem cell types (Figure 1B) could be one of the reasons for this observation. In order to test this, we analyzed the results of FAME and the HG test using the miRNA expression data in SCD. When we detected over-representation of miRNA targets in a cluster, we tested whether the miRNA expression pattern and the average expression pattern of the mRNA cluster were significantly anti-correlated (Figure 2C). Similarly, we tested cases of miRNA target depletion for a significant positive correlation. In cases where the miRNA family contained more than one miRNA with expression data in SCD, we chose as a representative the miRNA that had the highest absolute value of expression correlation with the cluster. We found that the most significant enrichments identified by FAME were consistently better supported by the miRNA expression data (Figure 3): FAME yielded evidence of a significant positive correlation of miRNAs and sets of genes depleted of their targets in 23% of the cases, and evidence of a negative correlation of miRNAs and their targets in 18% of the cases. These results suggest that miRNA target depletion is more effective than enrichment in identifying functionally relevant miRNAs using co-expression data. Indeed, as described below, for several miRNAs with a known function in specific differentiation-related processes, we found evidence of depletion of target sites in genes expressed during the same developmental stage, but no evidence of enrichment of target sites in genes expressed at other stages.Figure 3.


Towards computational prediction of microRNA function and activity.

Ulitsky I, Laurent LC, Shamir R - Nucleic Acids Res. (2010)

Performance of methods for enrichment detection on co-expression clusters. Out of the 1323 possible miRNA–cluster pairs, those with a correlation of r > 0.5 or r < −0.5 between the miRNA and the average mRNA expression were marked as ‘high’ (∼10% for each direction). The plots show the fraction of the 100 most significant miRNA–cluster pairs found by FAME and the HG test that fell into the ‘high’ category.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2926627&req=5

Figure 3: Performance of methods for enrichment detection on co-expression clusters. Out of the 1323 possible miRNA–cluster pairs, those with a correlation of r > 0.5 or r < −0.5 between the miRNA and the average mRNA expression were marked as ‘high’ (∼10% for each direction). The plots show the fraction of the 100 most significant miRNA–cluster pairs found by FAME and the HG test that fell into the ‘high’ category.
Mentions: According to the currently accepted model, miRNAs function as repressors of gene expression, and thus their expression patterns are expected to be anti-correlated with those of their targets (50,51) and correlated with their anti-targets [genes depleted of miRNA target sites (18)]. However, it is often difficult to identify such effects in matched miRNA/mRNA expression datasets. For example, in the SCD, the average Pearson correlation between expression of miRNAs and their TargetScan targets is 0.009. The uneven 3′ UTR lengths of genes highly expressed in different stem cell types (Figure 1B) could be one of the reasons for this observation. In order to test this, we analyzed the results of FAME and the HG test using the miRNA expression data in SCD. When we detected over-representation of miRNA targets in a cluster, we tested whether the miRNA expression pattern and the average expression pattern of the mRNA cluster were significantly anti-correlated (Figure 2C). Similarly, we tested cases of miRNA target depletion for a significant positive correlation. In cases where the miRNA family contained more than one miRNA with expression data in SCD, we chose as a representative the miRNA that had the highest absolute value of expression correlation with the cluster. We found that the most significant enrichments identified by FAME were consistently better supported by the miRNA expression data (Figure 3): FAME yielded evidence of a significant positive correlation of miRNAs and sets of genes depleted of their targets in 23% of the cases, and evidence of a negative correlation of miRNAs and their targets in 18% of the cases. These results suggest that miRNA target depletion is more effective than enrichment in identifying functionally relevant miRNAs using co-expression data. Indeed, as described below, for several miRNAs with a known function in specific differentiation-related processes, we found evidence of depletion of target sites in genes expressed during the same developmental stage, but no evidence of enrichment of target sites in genes expressed at other stages.Figure 3.

Bottom Line: Our analysis is based on a novel compendium of experimentally verified miRNA-pathway and miRNA-process associations that we constructed, which can be a useful resource by itself.Our method also predicts novel miRNA-regulated pathways, refines the annotation of miRNAs for which only crude functions are known, and assigns differential functions to miRNAs with closely related sequences.Applying our approach to groups of co-expressed genes allows us to identify miRNAs and genomic miRNA clusters with functional importance in specific stages of early human development.

View Article: PubMed Central - PubMed

Affiliation: Blavatnik School of Computer Science, Tel Aviv University, Tel Aviv, Israel. ulitsky@wi.mit.edu

ABSTRACT
While it has been established that microRNAs (miRNAs) play key roles throughout development and are dysregulated in many human pathologies, the specific processes and pathways regulated by individual miRNAs are mostly unknown. Here, we use computational target predictions in order to automatically infer the processes affected by human miRNAs. Our approach improves upon standard statistical tools by addressing specific characteristics of miRNA regulation. Our analysis is based on a novel compendium of experimentally verified miRNA-pathway and miRNA-process associations that we constructed, which can be a useful resource by itself. Our method also predicts novel miRNA-regulated pathways, refines the annotation of miRNAs for which only crude functions are known, and assigns differential functions to miRNAs with closely related sequences. Applying our approach to groups of co-expressed genes allows us to identify miRNAs and genomic miRNA clusters with functional importance in specific stages of early human development. A full list of the predicted mRNA functions is available at http://acgt.cs.tau.ac.il/fame/.

Show MeSH
Related in: MedlinePlus