RNAsnap2

RNAsnap2 predicts RNA solvent accessibility using deep learning models that integrate sequence-derived and structural features.

Key Features:

• Dilated Convolutional Neural Network: Implements a dilated convolutional neural network (CNN) architecture incorporating predicted base-pairing probabilities from LinearPartition to improve solvent accessibility prediction.
• Profile-Based and Single-Sequence Modes: Provides a profile-based model using homology-derived sequence profiles generated by Infernal and a single-sequence version (RNAsnap2 SingleSeq) that operates without sequence profiles while maintaining comparable performance.

Scientific Applications:

• Functional Region Identification: Identifies solvent-exposed regions and structural signatures in coding and non-coding RNAs, supporting analysis of RNA structure–function relationships.

Methodology:

RNAsnap2 applies a dilated CNN trained on RNA structural data, integrating sequence information and LinearPartition-derived base-pairing probabilities to predict per-nucleotide solvent accessibility, with optional homology-based profiles generated by Infernal.