EMDL_m6Am: identifying N6,2'-O-dimethyladenosine sites based on stacking ensemble deep learning

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• N6,2 -O-dimethyladenosine (m⁶Am) is a prevalent RNA modification in vertebrate mRNA's transcription initiation region.
• m⁶Am modification is linked to human diseases such as obesity and stomach cancer.
• Accurate identification of m⁶Am sites is crucial for understanding its regulatory roles in RNA.

Purpose of the Study:

• To develop a novel deep learning model for predicting m⁶Am sites in RNA sequences.
• To enhance the accuracy and efficiency of m⁶Am site identification.
• To provide a valuable tool for future research on m⁶Am's biological functions and disease associations.

Main Methods:

• Developed EMDL_m6Am, a deep learning model utilizing one-hot encoding for RNA sequence feature representation.
• Integrated multiple Convolutional Neural Network (CNN) models, including DenseNet, DCNN, and MSRN, through stacking.
• Evaluated model performance using sensitivity, specificity, accuracy, MCC, and AUC on training and independent test datasets.

Main Results:

• The EMDL_m6Am model achieved high performance on the training dataset: 86.62% sensitivity, 88.94% specificity, 87.78% accuracy, 0.7590 MCC, and 0.8778 AUC.
• On the independent test set, the model demonstrated strong predictive capabilities with 82.25% sensitivity, 79.72% specificity, 80.98% accuracy, 0.6199 MCC, and 0.8211 AUC.
• The deep learning approach significantly improved the prediction of m⁶Am sites compared to existing methods.

Conclusions:

• EMDL_m6Am significantly enhances the predictive performance for m⁶Am sites.
• The model serves as a valuable reference for future studies investigating m⁶Am.
• Source code and data are publicly available to facilitate further research and development.