TargetScore

TargetScore predicts microRNA (miRNA) targets by integrating genome-wide expression fold-changes from miRNA overexpression experiments with sequence-based features using a variational Bayesian Gaussian mixture model to provide probabilistic target scores.

Key Features:

• Probabilistic Scoring Methodology: Uses a variational Bayesian Gaussian mixture model (VB-GMM) to analyze log fold-changes and infer posterior probabilities of miRNA targets.
• Integration of Expression and Sequence Data: Combines genome-wide expression profiling from miRNA overexpression experiments with sequence-based scores to produce integrated target scores.
• High Accuracy and Confidence: Demonstrates higher accuracy than existing methods, with predicted targets showing comparable protein downregulation and significant enrichment for Gene Ontology terms.
• Extensive Dataset Utilization: Developed using 84 datasets from the Gene Expression Omnibus covering 77 human tissues or cells and 113 distinct transfected miRNAs.
• Research Applications: Applied to explore oncomir–oncogene networks and predict potential cancer-related miRNA–mRNA interactions.

Scientific Applications:

• Oncomir–oncogene network analysis: Prediction of miRNA–mRNA interactions implicated in cancer-related regulatory networks.
• Cross-tissue miRNA target identification: Identification of miRNA targets across diverse human tissues and cells using compiled GEO datasets.
• Functional and protein-level validation: Prioritization of targets that exhibit protein downregulation and Gene Ontology enrichment for downstream validation.

Methodology:

Integrates expression data from miRNA overexpression experiments with sequence-based scoring; applies VB-GMM to calculate posterior distributions of latent variables representing miRNA targets; computes final scores as a sigmoid-transformed fold-change weighted by averaged posteriors over all features.