Antar

Antar predicts microRNA (miRNA) targets in human and mouse genomes and discriminates binding determinants from mRNA-degradation-associated features using models trained on microarray expression (miRNA transfection/inhibition) and Argonaute CLIP-seq (HITS-CLIP, PAR-CLIP) data.

Key Features:

• Dual Predictive Models: Implements two complementary models trained on microarray expression data from miRNA transfection/inhibition and on Argonaute CLIP-seq datasets (HITS-CLIP and PAR-CLIP) to capture both repression and binding signals.
• Nonparametric Predictive Models: Uses nonparametric models to analyze a comprehensive set of known target features and flanking regions integrating expression and CLIP evidence.
• Spatial Analysis via CLIP-seq Data: Exploits CLIP-seq cross-linking positions to evaluate spatial effects of flanking features, including conservation around RNA-binding protein sites, on miRNP binding and cross-linking efficiency.
• Distinct Determinants for Target Prediction: Identifies that seed-related features predominate in expression-based datasets while flanking region conservation is a major AGO-binding determinant in CLIP-seq datasets.
• Independence of mRNA Degradation: Reports that CLIP-detected targets do not necessarily correlate with subsequent mRNA degradation levels, indicating separable binding and repression mechanisms.
• Predictive Power Across Datasets: Shows that CLIP-seq–trained models predict independent CLIP datasets robustly but do not predict expression changes, while expression-trained models poorly predict CLIP targets.

Scientific Applications:

• AGO-binding analysis: Elucidates initial miRNA targeting events by analyzing Argonaute binding locations from HITS-CLIP and PAR-CLIP.
• Determinant comparison: Characterizes the relative contributions of seed matches and flanking region conservation to target recognition and AGO binding.
• Decoupling binding and repression: Dissects the relationship between miRNA–target binding (CLIP signal) and downstream mRNA degradation observed in microarray transfection/inhibition studies.

Methodology:

Nonparametric predictive models are trained on microarray expression data (miRNA transfection/inhibition) and Argonaute CLIP-seq (HITS-CLIP, PAR-CLIP); models analyze known target features and flanking regions and use CLIP cross-linking positions to assess spatial determinants such as conservation adjacent to RNA-binding protein sites.