HydRA: Deep-learning models for predicting RNA-binding capacity from protein interaction association context and...

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• RNA-binding proteins (RBPs) are crucial regulators of gene expression, and their dysfunction is linked to human diseases.
• Identifying RBPs and their RNA-binding domains (RBDs) is essential for understanding cellular processes.
• Current methods struggle to identify RBPs lacking canonical RBDs, limiting comprehensive RBP cataloging.

Purpose of the Study:

• To develop a highly specific and sensitive computational tool for predicting RNA-binding capacity.
• To identify novel RNA-binding domains beyond canonical RBDs.
• To accelerate the discovery of a comprehensive catalog of RBPs and their functions.

Main Methods:

• Developed HydRA, a hybrid ensemble classifier integrating protein interaction networks and sequence patterns.
• Utilized machine learning models including Support Vector Machines (SVMs), Convolutional Neural Networks (CNNs), and Transformer-based protein language models.
• Employed occlusion mapping for domain detection and Enhanced CLIP (eCLIP) to validate RNA targets and binding activity.

Main Results:

• HydRA demonstrates unparalleled specificity and sensitivity in predicting RNA-binding capacity.
• Identified hundreds of uncharacterized RNA-binding associated domains.
• eCLIP experiments validated HydRA's predictions, confirming RNA targets and binding activity for novel domains.

Conclusions:

• HydRA significantly advances the identification of RBPs and their associated domains.
• Expands the known diversity of RNA-binding associated domains, crucial for understanding gene regulation.
• Provides a powerful tool for constructing comprehensive RBP catalogs and investigating RBP-related diseases.