rsRNASP

rsRNASP evaluates RNA three-dimensional (3D) structures using a residue-separation-based, all-atom distance-dependent statistical potential that distinguishes short- and long-range residue interactions to assess model accuracy.

Key Features:

• All-Atom Distance-Dependent Statistical Potential: rsRNASP employs an all-atom, distance-dependent statistical potential to model atomic interactions within RNA 3D structures.
• Residue Separation Distinction: Potentials are partitioned into short- and long-range terms based on residue separation distance to capture interactions at different spatial scales.
• Knowledge-Based Statistical Potential: Scoring functions are derived from knowledge-based statistical potentials specifically tailored for RNA.
• Performance on Large RNAs and Prediction Models: Extensive testing shows superior performance for evaluating large RNA structures derived from structure prediction models, with particularly high performance on the RNA-Puzzles dataset.
• Competitive on Small RNAs and Near-Native Decoys: Achieves performance comparable to top-performing statistical potentials for smaller RNAs and near-native decoy sets.
• Comparison to Neural Network-Based Scoring (RNA3DCNN): Demonstrated to outperform RNA3DCNN, a scoring function developed using 3D convolutional neural networks, without relying on neural-network black-box models.

Scientific Applications:

• RNA Structure Evaluation and Prediction: Scores and ranks predicted RNA 3D models to assess accuracy relative to experimental structures.
• Benchmarking and Validation: Serves to benchmark and validate RNA structure prediction methods and decoy sets, including evaluations using the RNA-Puzzles dataset.

Methodology:

Derives knowledge-based, all-atom distance-dependent statistical potentials partitioned by residue separation into short- and long-range terms, and applies these potentials to score decoys and benchmark predictions against RNA datasets (e.g., RNA-Puzzles) and to compare with RNA3DCNN.