Pprint2

Pprint2 predicts RNA-interacting residues in proteins using machine learning and deep learning approaches to identify protein–RNA interaction sites for studies of gene regulation and RNA processing.

Key Features:

• Dataset Utilization: Trained on 545 non-redundant RNA-binding proteins with an independent validation set of 161 proteins.
• Machine Learning Approach: Employs machine learning models using binary profiles with AUC=0.68 on the validation dataset, improved to AUC=0.76 by integrating evolutionary profiles.
• Convolutional Neural Networks: Incorporates convolutional neural networks which further increase predictive performance to AUC=0.82.
• Performance Metrics: Final model reports AUC of 0.82 and Matthews correlation coefficient (MCC) of 0.49 on the validation dataset, outperforming existing methods on that set.
• Biological Insights: Highlights enrichment of positively charged amino acids histidine (H), arginine (R), and lysine (K) among RNA-interacting residues.

Scientific Applications:

• Protein–RNA interaction mapping: Predicts residue-level RNA-binding sites to support studies of molecular mechanisms in protein–RNA interactions.
• Functional and therapeutic studies: Informs investigations of gene regulation, RNA processing, disease mechanisms, and development of therapeutic strategies targeting protein–RNA interfaces.

Methodology:

Trained on 545 non-redundant RNA-binding proteins with a 161-protein validation set, using machine learning models with binary profiles, integration of evolutionary profiles, and convolutional neural networks.