DeepPromise

DeepPromise predicts N1-methyladenosine (m1A) and N6-methyladenosine (m6A) RNA modification sites using deep learning to identify sequence motifs and positional nucleotide importance associated with these post-transcriptional modifications.

Key Features:

• Target modifications: Predicts N1-methyladenosine (m1A) and N6-methyladenosine (m6A) modification sites.
• Feature encodings: Integrates enhanced nucleic acid composition, one-hot encoding, and RNA embedding to represent sequence context.
• Model architecture: Processes encoded inputs through seven consecutive convolutional neural network (CNN) layers.
• Score integration: Combines prediction scores from CNN-based models to refine final predictions.
• Sequence positional analysis: Emphasizes the importance of proximal nucleotide positions while accounting for variable contributions from distal positions.
• Performance: Reports approximately 43% higher AUROC for m1A site prediction and a 2–6% increase in AUROC for m6A prediction on independent test datasets.
• Input types: Accepts RNA sequences and genomic sequences as input for prediction.

Scientific Applications:

• m1A site prediction: Identification and prioritization of candidate N1-methyladenosine modification sites in RNA sequences.
• m6A site prediction: Identification and prioritization of candidate N6-methyladenosine modification sites in RNA sequences.
• Motif characterization: Extraction and characterization of sequence motifs associated with m1A and m6A modifications.
• Complementing experimental methods: Prioritization of putative modification sites to complement high-throughput experimental assays.

Methodology:

Sequences are encoded using enhanced nucleic acid composition, one-hot encoding, and RNA embedding, processed through seven consecutive CNN layers, and their CNN prediction scores are combined to produce final site predictions.