MBSTAR

MBSTAR predicts functional microRNA (miRNA) binding sites in target mRNA 3' untranslated regions (3' UTRs) to distinguish functional from non-functional potential binding sites.

Key Features:

• Multiple instance learning: Applies multiple instance learning to manage uncertainty about the exact locations of true miRNA binding sites on target mRNAs.
• Training data (Tarbase 6.0): Trains on 9531 interacting miRNA–mRNA pairs and 973 non-interacting pairs derived from Tarbase 6.0.
• PAR-CLIP confirmation: Uses PAR-CLIP dataset overlap to confirm predicted binding sites.
• Functional-site prioritization: Distinguishes functional versus non-functional potential binding sites to reduce false positives in miRNA target prediction.
• Performance metric: Demonstrates prediction performance with a maximum F-Score of 0.337 and high overlap with PAR-CLIP validated sites.

Scientific Applications:

• Functional binding-site identification: Identification of functional miRNA binding sites in 3' UTRs for downstream analyses.
• Prioritization for experimental validation: Prioritizes candidate sites that overlap PAR-CLIP evidence for more reliable experimental follow-up.
• Genome-wide target analysis: Enables genome-wide assessment of putative functional miRNA–mRNA interactions.
• miRNA-mediated regulation studies: Supports investigation of gene regulation mechanisms mediated by miRNAs.

Methodology:

MBSTAR employs multiple instance learning trained on Tarbase 6.0 (9531 interacting pairs and 973 non-interacting pairs) and confirms predictions using PAR-CLIP overlap, with performance evaluated by overlap metrics and F-Score (maximum 0.337).