Limits...
Identifying TF-MiRNA Regulatory Relationships Using Multiple Features.

Shao M, Sun Y, Zhou S - PLoS ONE (2015)

Bottom Line: To further improve the accuracy of our inference, we also apply a mean reciprocal rank (MRR)-based method to the candidate peaks.We apply our pipeline to infer TF-miRNA regulatory relationships in mouse embryonic stem cells.The experimental results show that our approach provides very specific findings of TF-miRNA regulatory relationships.

View Article: PubMed Central - PubMed

Affiliation: School of Computer Science and Shanghai Key Lab of Intelligent Information Processing, Fudan University, 220 Handan Road, Shanghai 200433, China; Department of Computer Science and Engineering, Michigan State University, 428 S. Shaw Lane, East Lansing, 48824, USA.

ABSTRACT
MicroRNAs are known to play important roles in the transcriptional and post-transcriptional regulation of gene expression. While intensive research has been conducted to identify miRNAs and their target genes in various genomes, there is only limited knowledge about how microRNAs are regulated. In this study, we construct a pipeline that can infer the regulatory relationships between transcription factors and microRNAs from ChIP-Seq data with high confidence. In particular, after identifying candidate peaks from ChIP-Seq data, we formulate the inference as a PU learning (learning from only positive and unlabeled examples) problem. Multiple features including the statistical significance of the peaks, the location of the peaks, the transcription factor binding site motifs, and the evolutionary conservation are derived from peaks for training and prediction. To further improve the accuracy of our inference, we also apply a mean reciprocal rank (MRR)-based method to the candidate peaks. We apply our pipeline to infer TF-miRNA regulatory relationships in mouse embryonic stem cells. The experimental results show that our approach provides very specific findings of TF-miRNA regulatory relationships.

Show MeSH
The recall and removal rate of prediction on Oct4-miRNA relationships using only known regulation between Oct4 and miRNA genes.In each panel, the x-axis denotes the parameter p of SVMlight. The y-axis denotes the recall (left panel) and the removal rate (right panel) of the prediction, respectively. f10 means 10% of the known positive examples are put into the unlabeled data,and f30, ⋯, f90 mean 30%, ⋯, 90% of the know positive examples are put into the unlabeled data, respectively.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4414601&req=5

pone.0125156.g005: The recall and removal rate of prediction on Oct4-miRNA relationships using only known regulation between Oct4 and miRNA genes.In each panel, the x-axis denotes the parameter p of SVMlight. The y-axis denotes the recall (left panel) and the removal rate (right panel) of the prediction, respectively. f10 means 10% of the known positive examples are put into the unlabeled data,and f30, ⋯, f90 mean 30%, ⋯, 90% of the know positive examples are put into the unlabeled data, respectively.

Mentions: The performance of using only known regulation between Oct4 and protein-coding genes as the positive training data (Fig 4) is similar to the above result, especially when a small fraction of positives are selected as unlabeled data. However, as the known TF-miRNA relationships are too limited, the performance of using only known regulation between Oct4 and miRNA genes as the positive training data (Fig 5) is not satisfactory. This result indicates that we can still predict the regulatory relationships for miRNA genes when only protein-coding gene related peaks are available. Thus, for transcription factors that do not have known regulated miRNAs (e.g. Esrrb and Klf4), we used protein-coding gene related peaks only as positive samples in the following experiments.


Identifying TF-MiRNA Regulatory Relationships Using Multiple Features.

Shao M, Sun Y, Zhou S - PLoS ONE (2015)

The recall and removal rate of prediction on Oct4-miRNA relationships using only known regulation between Oct4 and miRNA genes.In each panel, the x-axis denotes the parameter p of SVMlight. The y-axis denotes the recall (left panel) and the removal rate (right panel) of the prediction, respectively. f10 means 10% of the known positive examples are put into the unlabeled data,and f30, ⋯, f90 mean 30%, ⋯, 90% of the know positive examples are put into the unlabeled data, respectively.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4414601&req=5

pone.0125156.g005: The recall and removal rate of prediction on Oct4-miRNA relationships using only known regulation between Oct4 and miRNA genes.In each panel, the x-axis denotes the parameter p of SVMlight. The y-axis denotes the recall (left panel) and the removal rate (right panel) of the prediction, respectively. f10 means 10% of the known positive examples are put into the unlabeled data,and f30, ⋯, f90 mean 30%, ⋯, 90% of the know positive examples are put into the unlabeled data, respectively.
Mentions: The performance of using only known regulation between Oct4 and protein-coding genes as the positive training data (Fig 4) is similar to the above result, especially when a small fraction of positives are selected as unlabeled data. However, as the known TF-miRNA relationships are too limited, the performance of using only known regulation between Oct4 and miRNA genes as the positive training data (Fig 5) is not satisfactory. This result indicates that we can still predict the regulatory relationships for miRNA genes when only protein-coding gene related peaks are available. Thus, for transcription factors that do not have known regulated miRNAs (e.g. Esrrb and Klf4), we used protein-coding gene related peaks only as positive samples in the following experiments.

Bottom Line: To further improve the accuracy of our inference, we also apply a mean reciprocal rank (MRR)-based method to the candidate peaks.We apply our pipeline to infer TF-miRNA regulatory relationships in mouse embryonic stem cells.The experimental results show that our approach provides very specific findings of TF-miRNA regulatory relationships.

View Article: PubMed Central - PubMed

Affiliation: School of Computer Science and Shanghai Key Lab of Intelligent Information Processing, Fudan University, 220 Handan Road, Shanghai 200433, China; Department of Computer Science and Engineering, Michigan State University, 428 S. Shaw Lane, East Lansing, 48824, USA.

ABSTRACT
MicroRNAs are known to play important roles in the transcriptional and post-transcriptional regulation of gene expression. While intensive research has been conducted to identify miRNAs and their target genes in various genomes, there is only limited knowledge about how microRNAs are regulated. In this study, we construct a pipeline that can infer the regulatory relationships between transcription factors and microRNAs from ChIP-Seq data with high confidence. In particular, after identifying candidate peaks from ChIP-Seq data, we formulate the inference as a PU learning (learning from only positive and unlabeled examples) problem. Multiple features including the statistical significance of the peaks, the location of the peaks, the transcription factor binding site motifs, and the evolutionary conservation are derived from peaks for training and prediction. To further improve the accuracy of our inference, we also apply a mean reciprocal rank (MRR)-based method to the candidate peaks. We apply our pipeline to infer TF-miRNA regulatory relationships in mouse embryonic stem cells. The experimental results show that our approach provides very specific findings of TF-miRNA regulatory relationships.

Show MeSH