PANTHER Score: Protein-Affinity for Nucleic Target-binding, Hybridization, and Energy Regression

Area of Science:

• Computational Biology
• Biophysics
• Machine Learning

Background:

• Protein-RNA interactions are vital for cellular functions.
• Accurate prediction of binding free energies (ΔG) is challenging due to data scarcity and interaction complexity.

Purpose of the Study:

• To develop a machine learning model, the PANTHER score, for predicting protein-RNA binding free energies.
• To address limitations in experimental data for protein-RNA interaction studies.

Main Methods:

• A local-to-global approach was used, deriving local interaction energies from molecular dynamics simulations.
• Machine learning models were trained to predict local interaction energies, integrated into the PANTHER score.
• The model was evaluated on test and external stress sets, including 110 complexes with experimental ΔG.

Main Results:

• Random Forest Regression achieved the highest predictive performance, yielding a Pearson correlation coefficient (r) of 0.80 and MAE of 1.79 kcal/mol on the test set.
• The model demonstrated strong predictive capabilities on the stress set (r=0.64, MAE=1.63 kcal/mol).
• The PANTHER score outperformed existing tools in benchmarking tests.

Conclusions:

• The PANTHER score is an effective tool for predicting protein-RNA binding affinities.
• Machine learning can overcome data limitations in predicting complex biomolecular interactions.
• This method advances biomolecular research and drug discovery by providing accurate binding energy predictions.