Deep4mC

Deep4mC predicts DNA N4-methylcytosine (4mC) sites in genomic sequences to support investigation of epigenetic regulation across multiple species.

Key Features:

• Deep learning-based prediction: Leverages convolutional neural networks (CNNs) to predict 4mC sites from sequence-derived features.
• Representative feature set: Incorporates four representative features that were systematically evaluated for predictive capacity across six species.
• Feature optimization and reinforcement learning: Employs feature selection strategies and reinforcement learning techniques to enhance model performance.
• Benchmark validation and performance metrics: Validated on a representative benchmark dataset achieving average area under curve (AUC) values greater than 0.9 across tested species, with AUC improvements of 10.14% to 46.21% over previous tools.
• Bootstrapping for small sample sizes: Extends the deep learning framework using a bootstrapping method to improve robustness for species with limited data.

Scientific Applications:

• Epigenetic research: Enables identification of putative 4mC sites to study epigenetic regulation of processes such as DNA replication, cell cycle regulation, and gene expression.
• Comparative genomics: Facilitates cross-species identification of potential 4mC modifications for comparative genomic analyses across the six evaluated species.
• Biomarker discovery: Supports discovery of candidate 4mC sites that may serve as biomarkers in disease research and diagnostics due to high predictive accuracy.

Methodology:

Convolutional neural networks trained on four representative features with feature selection and reinforcement learning, evaluated on a benchmark dataset using AUC metrics, and supplemented by bootstrapping for small-sample species.