Deepred-Mt

Deepred-Mt predicts cytidine (C) to uridine (U) RNA editing sites in angiosperm mitochondrial RNA to identify editing events that alter mitochondrial protein sequences and regulatory elements.

Key Features:

• Model architecture: A deep convolutional neural network evaluates sequence features to classify central cytidines as edited or unedited.
• Input window: The model uses a 40-nucleotide sequence segment with 20 nucleotides flanking each side of the central cytidine.
• Editing extent integration: Predictions incorporate editing extent information to improve discrimination of edited sites.
• Data augmentation: Novel data augmentation strategies are applied during training to enhance model generalization.
• Training data: The model is trained on deep RNA sequencing data from 21 plant mitochondrial genomes.
• Motif and structure recognition: The approach identifies established sequence motifs and highlights the potential regulatory role of local RNA structures around editing sites.
• Performance metrics: The method outperforms sequence homology–based approaches, achieving superior average precision and F1 scores.

Scientific Applications:

• Genetic engineering research: Provides precise predictions of C-to-U editing sites to support studies and interventions aimed at RNA-level regulation.
• Mechanistic studies of RNA editing: Offers insights into sequence motifs and local RNA structural factors implicated in editing mechanisms.
• Comparative analysis across angiosperms: Enables analysis of editing site patterns and prediction reliability across diverse angiosperm mitochondrial genomes.

Methodology:

Deep convolutional neural network classification on 40-nt windows centered on cytidines, integration of editing extent information, application of data augmentation strategies, and training on deep RNA sequencing data from 21 plant mitochondrial genomes with evaluation using average precision and F1 scores against homology-based methods.