Assessment of nucleic acid structure prediction in CASP16

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Accurate 3D structure prediction of nucleic acids (RNA and DNA) is crucial for understanding biological functions.
• The CASP16 (Critical Assessment of Structure Prediction) competition assessed the state-of-the-art in structure prediction.

Purpose of the Study:

• To evaluate the accuracy of computational methods for predicting 3D nucleic acid structures and complexes in a blind testing setting.
• To compare the performance of human expert groups versus automated servers in nucleic acid structure prediction.

Main Methods:

• Blind prediction of 42 diverse nucleic acid targets (monomers and complexes) in CASP16.
• Assessment of prediction accuracy using metrics like TM-scores.
• Analysis of factors influencing accuracy, including template availability and prediction methods.

Main Results:

• Overall performance in nucleic acid structure prediction was generally poor, with no novel RNA structures reaching high accuracy (TM-score > 0.8).
• Top-performing groups (Vfold, GuangzhouRNA-human, KiharaLab) were human expert predictors, indicating the continued importance of human expertise.
• Accuracy for nucleic acid complexes was also low unless 3D templates were available, similar to monomer prediction.
• Consistent recovery of complex structural features like pseudoknots and non-canonical pairs remained a challenge.

Conclusions:

• Despite improvements in automated servers, predicting novel 3D nucleic acid structures and complexes accurately remains a significant challenge.
• Prediction accuracy is highly dependent on the availability of homologous 3D structural templates.
• There has been no substantial increase in nucleic acid modeling accuracy in blind challenges between previous rounds and CASP16.