SPOT-RNA2

SPOT-RNA2 predicts RNA secondary structures, including pseudoknots and non-canonical base-pairs, using deep learning and transfer learning to improve modeling of long non-coding RNAs (lncRNAs) and other RNA molecules.

Key Features:

• Evolutionary Profile Integration: Incorporates evolutionary profiles from homologous sequences and achieves an F1-score >0.8 for 14 of 16 RNA samples when more than 1000 homologous sequences are available.
• Mutational Coupling: Employs mutational coupling data to refine identification of base-pairing interactions that contribute to tertiary structures.
• Two-dimensional Transfer Learning: Applies two-dimensional transfer learning using high-resolution 3-D structural data as a reference to improve prediction of canonical and non-canonical RNA secondary structures.
• Incorporation of Artificial Homologous Sequences: Integrates artificial homologous sequences generated from deep mutational scanning to boost predictive accuracy without modifying the trained model.
• Tertiary Base-pairing Prediction: Predicts tertiary base-pairing information, including pseudoknots and non-canonical base-pairs, to inform three-dimensional modeling.
• Automatic Prediction Capability: Produces fully automatic predictions of RNA secondary structures with tertiary base-pairing annotations.

Scientific Applications:

• RNA structure–function studies: Enables analysis of RNA structure–function relationships, including roles of non-coding RNAs such as lncRNAs.
• Three-dimensional modeling: Provides tertiary base-pairing constraints used to build three-dimensional RNA models.
• Drug design: Supplies structural information that can inform RNA-targeted drug design efforts.
• Functional annotation: Aids functional annotation of RNA molecules via predicted secondary and tertiary interactions.
• Evolutionary studies: Facilitates evolutionary analyses by leveraging homologous sequence information and evolutionary profiles.

Methodology:

Uses deep learning algorithms trained on large datasets of approximate RNA structures, applies two-dimensional transfer learning with high-resolution 3-D structural data, and integrates evolutionary profiles, mutational coupling data, and artificial homologous sequences from deep mutational scanning.