CoCoNet

CoCoNet combines direct coupling analysis (DCA) with a shallow convolutional neural network to improve RNA contact-map prediction and provide constraints for RNA structure modeling.

Key Features:

• Integration of Coevolutionary Models: CoCoNet employs direct coupling analysis (DCA) to extract coevolutionary signals for RNA contact prediction.
• Shallow Convolutional Neural Network: A shallow CNN with a limited number of parameters refines DCA outputs by learning spatial patterns in contact maps.
• Improved Accuracy: CoCoNet improves RNA contact-map prediction by approximately 70% over straightforward DCA, as validated by cross-validation on a dataset of around sixty RNA structures.
• Robustness and Generalization: The limited parameter set of the model promotes robustness and generalizability across diverse RNA datasets.

Scientific Applications:

• RNA Structure Prediction: More accurate contact maps from CoCoNet provide constraints that can be integrated into RNA tertiary structure modeling.
• Structural Biology Research: CoCoNet enables inference of RNA structural contacts in cases with sparse experimental data, informing studies of RNA function and interactions.

Methodology:

CoCoNet applies DCA to identify putative contacts from evolutionary information and then processes the DCA output through a shallow convolutional neural network to refine contact predictions.