Assessment of three-dimensional RNA structure prediction in CASP15.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Accurate prediction of RNA three-dimensional structures is a critical challenge in molecular biology.
• The CASP (Critical Assessment of protein Structure Prediction) competition has historically focused on protein structure prediction.

Purpose of the Study:

• To assess the state-of-the-art in RNA structure modeling through the first CASP RNA structure prediction exercise (CASP15).
• To evaluate the performance of various computational approaches, including deep learning, in predicting RNA three-dimensional structures.

Main Methods:

• Forty-two computational groups submitted RNA structure models for twelve RNA-containing targets in CASP15.
• Models were evaluated by RNA-Puzzles organizers and a CASP-recruited team using generalized protein assessment metrics (GDT, lDDT) and Z-score rankings.
• Predictions were further validated by comparison with experimental data, including cryogenic electron microscopy (cryo-EM) maps and X-ray diffraction data.

Main Results:

• Two independent assessments identified AIchemy_RNA2, Chen, and RNAPolis/GeneSilico as the top-performing groups.
• Top-ranked groups, which did not utilize deep learning, significantly outperformed deep learning approaches.
• CASP15 models generally predicted the global fold of RNA targets accurately, with exceptions for two RNA-protein complexes.

Conclusions:

• Current RNA modeling approaches show promise for applications in RNA nanotechnology and structural biology.
• Despite successes in predicting global folds, challenges persist in accurately modeling fine structural details, such as noncanonical base pairs and multiple conformations.
• Further development is needed to improve the ranking of submitted models and the prediction of intricate structural features.