RNAPosers

RNAPosers classifies RNA–small molecule docking poses using machine learning to identify native-like 3D RNA–ligand complexes.

Key Features:

• Pose Fingerprinting: The pose fingerprint (FP) captures the local RNA environment around ligand atoms via composite atomic fingerprints to represent RNA–ligand interactions.
• Machine Learning Classifiers: Classifiers are trained to estimate the relative "nativeness" of RNA–ligand poses and to rank poses by likelihood of being native-like.
• Implementation: The codebase uses C++ and Python, with the pose fingerprint module compiled as an executable (bin/featurize) and prediction scripts compatible with Python 2 and Python 3 and multiple scikit-learn versions.
• Empirical Performance: In a leave-one-out evaluation on an 80-instance dataset, a classifier identified poses within 2.5 Å of native structures in approximately 80% of cases and achieved similar success rates on two independent validation sets.

Scientific Applications:

• RNA–Small Molecule Structure Prediction: Improves recovery of native-like RNA–ligand poses from docking predictions to support 3D structure determination.
• Drug Discovery: Prioritizes native-like poses to inform small-molecule design and selection targeting RNA.
• Molecular Recognition Studies: Enables analysis of precise RNA–ligand interactions to elucidate molecular recognition mechanisms in molecular biology.

Methodology:

The method uses a two-step computational workflow: generation of pose fingerprints encoding RNA–ligand interaction features, followed by machine learning classification of those fingerprints to distinguish native-like from non-native poses based on nativeness scores.