Nucpred

Nucpred predicts protein-binding nucleotides in RNA sequences using machine learning classifiers trained on nucleotide-triplet and nucleotide-quartet features to identify RNA–protein interaction sites.

Key Features:

• Machine learning classifiers: Comparative analysis of various machine learning classifiers was used to identify protein-binding nucleotides in RNA sequences.
• Feature encoding: Predictions use nucleotide-triplet and nucleotide-quartet sequence features optimized for performance.
• Model architecture: Random forest models trained on nucleotide-triplet and nucleotide-quartet features were found to be particularly effective.
• Performance metrics: Reported performance includes accuracy 84.8%, sensitivity 83.2%, specificity 86.1%, Matthews Correlation Coefficient 0.70, and AUC 0.93.
• Training data: Models were trained using high-throughput sequencing data to compensate for limited structural data on RNA–protein complexes.

Scientific Applications:

• RNA–protein interaction mapping: Predicting protein-binding nucleotides to advance the study of RNA–protein interactions.
• Mechanistic insight: Inferring binding interfaces to provide mechanistic insights into molecular functioning and aberrations.
• Disease biology: Supporting analysis of RNA-related disease mechanisms by identifying potential altered RNA–protein interactions.
• Complementing structural studies: Providing predictions where experimental co-crystallized RNA–protein complex structures are scarce and RNA flexibility complicates structural determination.

Methodology:

Models were trained on nucleotide sequence features derived from high-throughput sequencing data via comparative evaluation of machine learning classifiers, with random forest models based on nucleotide-triplet and nucleotide-quartet features demonstrating strong performance.