DL-m6A

DL-m6A predicts N6-methyladenosine (m6A) sites in mammalian RNA sequences using deep learning to improve transcriptome-wide and tissue-specific m6A site identification.

Key Features:

• Deep learning model: Implements a deep learning-based classifier to predict N6-methyladenosine (m6A) sites.
• Three encoding schemes: Uses three distinct encoding schemes to generate comprehensive contextual feature representations of input RNA sequences.
• Shallow feature extraction: Extracts shallow features from individual contextual vectors using several neural network layers.
• Feature concatenation: Concatenates individual contextual vectors into a unified feature vector for integrated representation.
• Deep feature extraction: Processes the unified feature vector through additional neural network layers to capture deeper sequence features for prediction.
• Evaluation datasets: Evaluated on tissue-specific and full transcript datasets to assess robustness and generalizability.
• Comparative performance: Demonstrated superior predictive performance relative to existing methods, including maintained accuracy when trained on full-transcript data and tested on tissue-specific data.

Scientific Applications:

• m6A site prediction: Prediction of N6-methyladenosine (m6A) sites in mammalian RNA sequences.
• Transcriptome-wide and tissue-specific profiling: Transcriptome-wide and tissue-specific m6A profiling and analysis.
• Method benchmarking: Benchmarking and comparison of computational m6A prediction methods.
• RNA modification studies: Supporting research into the biological implications of RNA modifications by providing predicted m6A site maps.

Methodology:

Input RNA sequences are encoded by three distinct schemes into contextual vectors; shallow features are extracted via several neural network layers, the vectors are concatenated into a unified feature vector, and additional neural network layers perform deeper feature extraction for final prediction.