RNAPosers: Machine Learning Classifiers for Ribonucleic Acid-Ligand Poses.

Area of Science:

• Computational biology
• Structural biology
• Bioinformatics

Background:

• Determining three-dimensional (3D) structures of ribonucleic acid (RNA)-small molecule ligand complexes is crucial for understanding molecular recognition.
• Traditional computer docking methods often misclassify poses due to limitations in scoring functions.

Purpose of the Study:

• To develop and evaluate machine learning-based pose classifiers for improved RNA-ligand 3D structure prediction.
• To introduce RNAPosers, a novel tool for enhancing RNA-ligand pose prediction accuracy.

Main Methods:

• Utilized machine learning to train pose classifiers using a composite pose fingerprint (FP) encoding local RNA environments.
• Employed a leave-one-out cross-validation approach for training and testing.
• Evaluated performance on two independent validation sets.

Main Results:

• Machine learning classifiers significantly outperformed traditional docking scores in recovering native-like RNA-ligand poses.
• One classifier achieved approximately 80% accuracy in recovering poses within 2.5 Å of native structures in 80 cases.
• Validation sets showed approximately 60% accuracy in recovering near-native poses.

Conclusions:

• RNAPosers provide a more accurate method for predicting RNA-ligand complex structures compared to existing docking scoring functions.
• The developed classifiers offer a valuable tool for researchers in RNA structural biology and drug discovery.
• RNAPosers are made publicly available to the academic community.