Deep-RBPPred: Predicting RNA binding proteins in the proteome scale based on deep learning.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• RNA binding proteins (RBPs) are crucial for cellular functions.
• Accurate identification of RBPs is essential for biological research.
• Previous methods like RBPPred are computationally intensive due to PSSM matrix generation.

Purpose of the Study:

• To develop a faster and accurate computational model for RBP prediction.
• To leverage deep learning, specifically Convolutional Neural Networks (CNNs), for RBP identification.
• To compare the performance of the new model against existing methods and different training strategies.

Main Methods:

• Developed Deep-RBPPred, a CNN-based model utilizing physicochemical properties from protein sequences.
• Trained two versions: Deep-RBPPred-balance and Deep-RBPPred-imbalance.
• Evaluated performance using Matthews Correlation Coefficient (MCC) on proteomes from A. thaliana, S. cerevisiae, and H. sapiens.
• Compared CNN with Support Vector Machine (SVM) algorithms.

Main Results:

• Deep-RBPPred demonstrates significant speed improvements and good generalization ability.
• Achieved high MCC values (e.g., 0.82 for balance model in A. thaliana).
• CNN-based models outperformed SVM, identifying more RBPs.
• Identified a substantial number of new RBPs (280 for balance, 265 for imbalance).
• The balance model showed ~7% higher sensitivity than state-of-the-art methods.
• Eukaryotic proteomes show higher RBP rates than bacterial proteomes.

Conclusions:

• Deep-RBPPred offers a computationally efficient and accurate alternative for RBP prediction.
• CNNs are effective for RBP identification, outperforming traditional machine learning methods like SVM.
• The study highlights the prevalence of RBPs in eukaryotes compared to bacteria.