Influence of contact map topology on RNA structure prediction

Area of Science:

• Computational biology
• Structural biology
• Bioinformatics

Background:

• Advances in RNA sequencing have generated vast amounts of data, but computational RNA structure prediction quality lags behind experimental methods.
• While AI models like AlphaFold3 can predict RNA structures, their confidence, especially for novel sequences, remains a limitation.
• Current two-step prediction strategies rely on contact maps, but their quality assessment has been incomplete.

Purpose of the Study:

• To investigate the impact of contact map properties on RNA structure prediction accuracy.
• To propose an improved method for evaluating contact maps beyond simple true positive proportion.
• To introduce a new metric, the Gaussian score, to quantify contact distribution.

Main Methods:

• Analyzing the relationship between contact map topology and RNA structure prediction quality.
• Evaluating the effect of contact clustering versus distribution on prediction accuracy using computational experiments.
• Developing and applying the Gaussian score to assess contact distribution in RNA contact maps.

Main Results:

• Contact map clustering, common in AI predictions, negatively impacts RNA structure prediction quality.
• A more distributed contact topology significantly improves prediction accuracy.
• The Gaussian score effectively measures contact distribution and correlates with improved prediction outcomes.

Conclusions:

• Contact map quality for RNA structure prediction depends on both the proportion and distribution of contacts.
• The Gaussian score offers a valuable metric for enhancing RNA structure prediction by favoring distributed contact topologies.
• These findings can guide the development of more accurate computational RNA structure prediction algorithms.