Predicting RNA SHAPE scores with deep learning.

Area of Science:

• Computational biology
• Molecular biology
• Bioinformatics

Background:

• Current RNA secondary structure prediction models rely on in vitro physical chemistry.
• In vivo RNA structure experiments using SHAPE-MaP offer new insights.
• Integrating in vivo data can improve computational RNA folding simulations.

Purpose of the Study:

• To develop a machine learning approach for predicting in vivo RNA structure.
• To enhance computational RNA folding accuracy by incorporating in vivo data.
• To predict in vivo SHAPE scores using RNA sequence and structure predictions.

Main Methods:

• Utilized a machine learning model.
• Input features included RNA secondary structure prediction results and nucleotide sequence.
• Predicted in vivo SHAPE scores.

Main Results:

• The machine learning approach achieved a higher Pearson correlation coefficient with experimental SHAPE scores compared to thermodynamic folding.
• Demonstrated improved accuracy in predicting in vivo RNA secondary structure.

Conclusions:

• Machine learning can effectively predict in vivo RNA SHAPE scores.
• This approach offers a significant improvement over traditional thermodynamic models for in vivo RNA structure prediction.
• Augmenting experimental data with computational predictions that consider in vivo folding properties is a promising direction.