DMISO

DMISO predicts miRNA and isomiR target sites on mRNAs using deep learning to improve base-pair resolution identification of miRNA-mRNA interactions.

Key Features:

• Deep learning model: Employs a deep learning approach to capture complex interaction patterns between miRNAs/isomiRs and mRNAs.
• IsomiR integration: Explicitly incorporates isomiR-mRNA interactions alongside canonical miRNA interactions in predictions.
• Base-pair resolution: Aims to identify miRNA/isomiR target sites at base-pair resolution.
• Interaction feature capture: Captures intricate interaction features between miRNA/isomiR and mRNA molecules beyond seed regions.
• Cross-validation: Performance assessed using tenfold cross-validation.
• Benchmark performance: Reported precision of 95% and recall of 90% and shown to outperform five existing tools (three conventional methods and two deep learning-based approaches) on three independent datasets.
• Feature interpretation: Applies feature-interpretation strategies to highlight the importance of miRNA regions beyond the seed and potential RNA-binding motifs in miRNAs/isomiRs and mRNAs.

Scientific Applications:

• miRNA/isomiR target prediction: Predicts miRNA and isomiR target sites on mRNAs at improved base-pair resolution.
• Method benchmarking: Provides comparative performance assessment against conventional and deep learning-based target prediction methods.
• Regulatory mechanism analysis: Identifies non-seed miRNA regions and RNA-binding motifs relevant to miRNA/isomiR-mediated gene regulation.

Methodology:

Uses a deep learning model, validated by tenfold cross-validation, evaluated on three independent datasets against five existing tools, and applies feature-interpretation strategies to identify important miRNA regions and RNA-binding motifs.