Direct Information Reweighted by Contact Templates (DIRECT)

DIRECT predicts RNA tertiary nucleotide-nucleotide contacts by integrating sequence co-variation with structural contact templates using a Restricted Boltzmann Machine.

Key Features:

• Contact template integration: DIRECT integrates structural contact templates with sequence co-variation data to inform contact predictions.
• Restricted Boltzmann Machine reweighting: A Restricted Boltzmann Machine (RBM) reweights direct information to enhance inference of nucleotide contacts.
• Benchmark performance: Benchmark tests report an average increase in prediction accuracy of 41% relative to mfDCA and 18% relative to plmDCA.
• Long-range and tertiary contact detection: The method improves prediction of long-range contacts and captures tertiary structural features.
• Hybrid approach: DIRECT combines machine learning with structural templates to augment traditional direct coupling analysis (DCA).

Scientific Applications:

• RNA structure-function analysis: Interpreting RNA structure-function relationships by predicting tertiary interactions.
• Structural biology: Informing structural biology studies of RNA by providing tertiary contact predictions.
• Genomics: Supporting genomics analyses that require accurate RNA structural information.
• Drug design: Enabling drug design efforts that target RNA by improving tertiary contact prediction.

Methodology:

DIRECT reweights direct information derived from sequence co-variation using a Restricted Boltzmann Machine trained on structural contact templates and is benchmarked against mfDCA and plmDCA.