RNAcontact

RNAcontact predicts inter-nucleotide 3D closeness in RNA sequences to inform RNA tertiary structure modeling.

Key Features:

• Deep Residual Neural Networks: Employs deep residual neural networks (ResNets) to predict RNA inter-nucleotide contacts.
• Covariance from Multiple Sequence Alignments: Utilizes covariance information derived from multiple sequence alignments to capture evolutionary constraints relevant to nucleotide proximity.
• Predicted Secondary Structure: Incorporates predicted secondary structures as input features to provide context beyond base pairing.
• Performance Metrics: Achieves precisions of 0.8 and 0.6 for the top L/10 and L predictions (where L is RNA length) on an independent test set, substantially outperforming traditional evolutionary coupling methods.
• Novel Predictions: Identifies interactions such that about one-third of correctly predicted 3D closenesses are not secondary-structure base pairs, revealing tertiary contacts beyond secondary structure.

Scientific Applications:

• RNA Structure Determination: Predicted inter-nucleotide closeness can be used as distance restraints to guide RNA folding with computational tools such as the 3dRNA package.
• Model Accuracy Improvement: Integrating predicted 3D closeness into modeling workflows yields more accurate RNA structural models compared to modeling without these restraints.

Methodology:

Trains deep residual neural networks on datasets comprising covariance data from multiple sequence alignments and predicted secondary structures.