SRCPred

SRCPred predicts RNA dinucleotide contacts within proteins by combining Position-Specific Scoring Matrices (PSSMs) and neural networks to identify protein–RNA interaction preferences.

Key Features:

• Evolutionary profiles (PSSMs): Represents protein subsequences as Position-Specific Scoring Matrices (PSSMs) to capture evolutionary information relevant to RNA binding.
• Neural network-based prediction: Employs neural networks trained on known protein–RNA complex structures to infer interaction preferences between protein subsequences and dinucleotides.
• Dinucleotide contact prediction: Predicts contacts at the dinucleotide level to identify regions enriched in specific dinucleotide interactions.
• Multiclass target vectors: Produces multiclass outputs for 16 possible contacting dinucleotide subsequences as a 16-dimensional contact probability matrix.
• Amino acid–dinucleotide contact statistics: Utilizes amino acid–dinucleotide contact statistics derived from protein–RNA complex structures to characterize pairing preferences.
• Cross-validation and accuracy: Reports cross-validation performance measured as area under the ROC curve (AUC) in the 65–80% range.

Scientific Applications:

• Dinucleotide-specific contact mapping: Maps dinucleotide-specific protein–RNA contacts to localize RNA-binding regions at dinucleotide resolution.
• Prediction of RNA-binding protein targets: Predicts potential RNA targets and interacting protein–RNA fragment pairs for RNA-binding proteins.
• Viral protein–RNA interaction analysis: Supports analysis of viral protein–RNA interactions relevant to replication, assembly, and antiviral strategy development.
• Molecular biology and bioinformatics studies: Provides dinucleotide-level interaction data useful in molecular biology, virology, and bioinformatics research.

Methodology:

Analyzes amino acid–dinucleotide contact statistics from protein–RNA complex structures, represents protein subsequences with PSSMs (evolutionary profiles), and trains neural networks to predict contact probabilities for 16 dinucleotide classes, with performance assessed by cross-validation (AUC).