DeepOMe

DeepOMe predicts 2'-O-methylation (2'-O-Me or Nm) sites within RNA sequences across the human transcriptome to support study of post-transcriptional regulation of gene expression.

Key Features:

• Target modification: Predicts 2'-O-methylation (2'-O-Me or Nm) sites in RNA sequences.
• Scope: Applies predictions across the human transcriptome.
• Architecture: Employs a hybrid deep-learning architecture integrating Convolutional Neural Networks (CNN) with Bidirectional Long Short-term Memory networks (BLSTM) for feature extraction and capturing long-range dependencies.
• Validation: Validated using 4-, 6-, 8-, and 10-fold cross-validation.
• Performance: Reports an Area Under the Curve (AUC) close to 0.998 and an Average Precision Recall (AUPR) near 0.880.
• Benchmarking: Evaluated on independent datasets and shown to outperform existing methods such as NmSEER V2.0.
• Context with experimental methods: Provides a computational alternative to experimental Nm-seq, which can have high costs and technical complexities.

Scientific Applications:

• Large-scale site detection: Enables large-scale detection of 2'-O-Me sites across transcripts.
• Regulatory studies: Supports investigation of the roles of 2'-O-methylation in post-transcriptional regulation and gene expression.
• Method comparison: Serves as a benchmark for comparing computational prediction methods against tools such as NmSEER V2.0 and experimental data from Nm-seq.

Methodology:

Uses a hybrid CNN and Bidirectional Long Short-term Memory (BLSTM) deep-learning architecture and is evaluated by 4-, 6-, 8-, and 10-fold cross-validation as well as independent dataset testing, with performance reported by AUC and AUPR.