KNetFold

KNetFold predicts consensus RNA secondary structures from alignments of RNA sequences using a hierarchical network of k-nearest neighbor classifiers that integrate mutual information, fraction of complementary nucleotides, and RNAfold consensus predictions.

Key Features:

• Hierarchical network of classifiers: A hierarchical network of k-nearest neighbor classifiers evaluates relationships between alignment positions to inform structure prediction.
• Input parameters: Uses mutual information, fraction of complementary nucleotides, and consensus RNAfold predictions for each pair of alignment columns and their nearest neighbors.
• Prediction output: Classifies each pair of alignment columns as "base pair" or "no base pair" to derive consensus secondary structures.
• Benchmark performance: Evaluated on 49 RFAM alignments with an average Matthews correlation coefficient (MCC) of 0.81, reported as superior to PFOLD and RNAalifold.

Scientific Applications:

• Consensus structure prediction: Determination of consensus secondary structures across aligned RNA sequences for comparative genomics analyses.
• Pseudoknot interaction prediction: Identification of pseudoknot interactions, demonstrated in applications to archaeal RNase P.

Methodology:

KNetFold applies a hierarchical network of k-nearest neighbor classifiers that integrate mutual information, fraction of complementary nucleotides, and consensus RNAfold predictions to classify each pair of alignment columns as "base pair" or "no base pair".