SPOT-Struct-RNA

SPOT-Struct-RNA predicts RNA-binding proteins (RBPs) and their RNA-binding amino acid residues from three-dimensional protein structures using structural alignment and statistical energy functions to identify protein–RNA interactions.

Key Features:

• Simultaneous prediction: Predicts both RBPs and RNA-binding amino acid residues by aligning query structures to known protein-RNA complex structures.
• Performance metrics: Achieves 98% accuracy and 91% precision for RBP prediction, and 93% accuracy and 78% precision for RNA-binding residue identification in leave-one-out cross-validation on a benchmark set of 212 RNA binding domains and 6761 non-RNA binding domains.
• Robust validation: Shows no false positives among 311 DNA binding domains, identifies six domains that bind both DNA and RNA, and correctly detected 31 of 75 unbound RNA-binding domains with 92% accuracy and 65% precision.
• Application to structural genomics: Applied to 2076 structural genomics targets, predicting 25 potential RBPs of which 80% were validated as putative RNA binders.
• Methodological components: Divides protein structures into domains, employs a Z-score for relative structural similarity, and uses a DFIRE-based statistical energy function to assess protein-RNA binding affinity.

Scientific Applications:

• Mechanistic studies of RBPs: Enables elucidation of RBP roles in gene regulation and RNA processing by predicting binding proteins and their interaction sites.
• Discrimination of nucleic acid binding: Differentiates RNA-binding domains from DNA-binding domains and identifies multifunctional domains that bind both RNA and DNA.
• Structural genomics annotation: Prioritizes putative RBPs among structural genomics targets for experimental validation.

Methodology:

Divide protein structures into domains; align structures to known protein-RNA complex structures; compute a Z-score for relative structural similarity; evaluate protein-RNA interactions with a DFIRE-based statistical energy function; and assess performance via leave-one-out cross-validation.