Improved prediction of DNA and RNA binding proteins with deep learning models

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Nucleic acid-binding proteins (NABPs), encompassing DNA-binding proteins (DBPs) and RNA-binding proteins (RBPs), are crucial for fundamental biological processes.
• Existing machine learning methods for NABP prediction face limitations due to dataset constraints and prediction scope.

Purpose of the Study:

• To develop more accurate and robust datasets for training and testing NABP prediction models.
• To create advanced deep learning models for predicting NABP types, overcoming limitations of prior approaches.

Main Methods:

• Generation of refined datasets for NABP classification.
• Development of deep learning models utilizing convolutional neural networks (CNNs) and long short-term memory (LSTM) networks.
• Implementation of both hierarchical and multi-class prediction strategies.

Main Results:

• The developed deep learning approaches demonstrated superior performance compared to existing DBP and RBP predictors.
• A balanced prediction accuracy was achieved between DBPs and RBPs, enhancing practical applicability.
• The multi-class approach significantly improved prediction accuracy, particularly for DBPs (~12% enhancement).
• Exploration of single-stranded DNA binding protein prediction accuracy and its impact on overall NABP prediction.

Conclusions:

• The novel deep learning strategies offer a more accurate and robust method for predicting nucleic acid-binding protein types.
• These advancements facilitate functional annotation and the identification of novel NABPs.
• The study highlights the effectiveness of multi-class deep learning for precise DBP and RBP classification.