N6-methyladenine identification using deep learning and discriminative feature integration.

Area of Science:

• Epigenetics and Genomics
• Computational Biology
• Bioinformatics

Background:

• N6-methyladenine (6 mA) is a key DNA modification regulating gene expression and biological processes.
• Accurate 6 mA site identification is crucial for early disease detection and understanding epigenetic mechanisms.
• Existing machine learning methods for 6 mA detection face challenges in generalizability and efficiency.

Purpose of the Study:

• To develop a novel Deep Neural Network (DNN) model, Deep-N6mA, for precise identification of 6 mA sites.
• To enhance the accuracy and efficiency of 6 mA site detection using hybrid feature extraction and selection.
• To improve the understanding of 6 mA's biological significance through advanced computational methods.

Main Methods:

• Utilized a Deep Neural Network (DNN) architecture for classification.
• Employed hybrid feature extraction including k-mer, Dinucleotide-based Cross Covariance (DCC), Trinucleotide-based Auto Covariance (TAC), Pseudo Single Nucleotide Composition (PseSNC), Pseudo Dinucleotide Composition (PseDNC), and Pseudo Trinucleotide Composition (PseTNC).
• Applied Principal Component Analysis (PCA) for unsupervised feature selection to optimize computational efficiency and informative feature extraction.

Main Results:

• Deep-N6mA achieved high accuracy: 97.70% on the F. vesca dataset and 95.75% on the R. chinensis dataset.
• The model outperformed existing methods by 4.12% and 4.55% on the respective datasets.
• Fivefold cross-validation demonstrated the robustness and generalizability of the Deep-N6mA model.

Conclusions:

• Deep-N6mA is a reliable and effective tool for precise N6-methyladenine site identification.
• The model contributes to advancing epigenetic research and computational biology.
• Deep-N6mA facilitates early detection and deeper understanding of 6 mA's role in biological systems.