lociPARSE: A Locality-aware Invariant Point Attention Model for Scoring RNA 3D Structures

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning Applications in Biochemistry

Background:

• Accurate assessment of 3D RNA structural models is crucial for evaluation, selection, and conformational sampling.
• Existing knowledge-based statistical potentials and machine learning methods struggle with high-fidelity RNA scoring.
• A reliable scoring function is needed for RNA structure prediction in the absence of experimental data.

Purpose of the Study:

• To introduce lociPARSE, a novel locality-aware invariant point attention architecture for scoring RNA 3D structures.
• To develop a method that accurately assesses RNA structural models without relying on experimental structures.
• To improve upon existing machine learning and statistical potential-based RNA scoring approaches.

Main Methods:

• Developed lociPARSE, a locality-aware invariant point attention architecture.
• Implemented a superposition-free approach to estimate Local Distance Difference Test (lDDT) scores.
• Aggregated local atomic environment accuracy to predict global structural accuracy for RNA models.

Main Results:

• lociPARSE significantly outperforms conventional statistical potentials (rsRNASP, cgRNASP, DFIRE-RNA, RASP) and machine learning methods (ARES, RNA3DCNN).
• Performance was validated across multiple datasets, including CASP15.
• lociPARSE demonstrates superior accuracy in assessing RNA 3D structural models.

Conclusions:

• lociPARSE provides a highly accurate and reliable method for scoring 3D RNA structures.
• The superposition-free lDDT estimation approach enhances local and global accuracy assessment.
• lociPARSE is a valuable tool for RNA structure evaluation, selection, and conformational sampling.