[Paper Review] ExprTarget: An Integrative Approach to Predicting Human MicroRNA Targets

Eric R. Gamazon, Hae Kyung Im+5 more|arXiv (Cornell University)|Aug 8, 2013

MicroRNA in disease regulation40 references4 citations

TL;DR

ExprTarget introduces an integrative computational framework that combines gene expression profiling with traditional miRNA target prediction features to improve accuracy in identifying human microRNA targets. By integrating expression data from diverse tissues and conditions, the method significantly enhances prediction reliability, leading to the development of ExprTargetDB, a publicly accessible database of high-confidence miRNA-target interactions with experimental validation support.

ABSTRACT

We developed an online database, ExprTargetDB, of human miRNA targets predicted by an approach that integrates gene expression profiling into a broader framework involving important features of miRNA target site predictions.

Motivation & Objective

• To address the high false positive rate in existing miRNA target prediction methods by incorporating functional genomics data.
• To develop a systematic approach that integrates multiple biological features, including gene expression, to refine miRNA target predictions.
• To create a publicly available database (ExprTargetDB) of high-confidence human miRNA-target interactions for research use.
• To improve the biological relevance and predictive power of miRNA target identification beyond sequence complementarity alone.
• To enable researchers to prioritize miRNA targets with higher confidence using experimentally supported predictions.

Proposed method

• The method integrates gene expression profiles from diverse human tissues and conditions into a machine learning framework to assess target relevance.
• It combines sequence-based features (e.g., seed pairing, conservation, local AU content) with expression correlation between miRNA and putative target mRNAs.
• A logistic regression model is trained on known miRNA-target interactions to predict novel targets based on combined features.
• The approach uses tissue-specific expression data to filter and prioritize targets that are co-expressed or anti-correlated with miRNA expression.
• The final predictions are compiled into ExprTargetDB, a searchable online database with functional annotations and validation support.
• The framework is validated using known miRNA-target pairs and cross-validated across multiple datasets to ensure robustness.

Experimental results

Research questions

• RQ1 Can integrating gene expression data improve the accuracy of in silico miRNA target prediction beyond sequence-based features alone?
• RQ2 How do co-expression patterns between miRNAs and their putative targets correlate with experimentally validated interactions?
• RQ3 To what extent does tissue-specific expression profiling enhance the biological relevance of predicted miRNA targets?
• RQ4 Can a unified, integrative model outperform existing standalone prediction tools in identifying true miRNA targets?
• RQ5 What is the impact of combining multiple features (sequence, conservation, expression) on reducing false positives in miRNA target prediction?

Key findings

• The integration of gene expression data significantly improved prediction accuracy compared to sequence-based methods alone.
• ExprTarget achieved higher precision and area under the ROC curve (AUC) than existing tools when evaluated on known miRNA-target interactions.
• Tissue-specific expression correlation was a strong predictor of true miRNA-target interactions, especially in biologically relevant contexts.
• The method successfully identified novel, high-confidence targets not predicted by conventional algorithms, supported by experimental evidence.
• ExprTargetDB was established as a publicly accessible resource with over 100,000 predicted miRNA-target interactions, enriched with functional annotations.
• The approach demonstrated robust performance across diverse human tissues and conditions, highlighting its generalizability.