Prediction of RNA-protein interactions by combining deep convolutional neural network with feature selection ensemble...

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• RNA-protein interactions (RPI) are crucial for fundamental cellular functions.
• Experimental determination of RPI is time-consuming and costly.
• There is a need for efficient computational methods to predict RPI.

Purpose of the Study:

• To develop a novel computational method, RPIFSE, for predicting RNA-protein interactions.
• To leverage RNA and protein sequence information for accurate RPI prediction.

Main Methods:

• RPIFSE utilizes a Feature Selection Ensemble approach.
• It combines Convolutional Neural Network (CNN) for feature extraction and Extreme Learning Machine (ELM) for classification.
• A weighted voting method integrates results from multiple classifiers.

Main Results:

• RPIFSE achieved high accuracy in 5-fold cross-validation: 91.87% (RPI369), 89.74% (RPI2241), 97.76% (RPI488), and 98.98% (RPI1807).
• Performance was superior to Support Vector Machine (SVM) and other existing methods.
• Successful prediction on the independent NPInter2.0 dataset and network graph generation demonstrated RPIFSE's utility.

Conclusions:

• RPIFSE is an effective computational tool for predicting RNA-protein interactions.
• The method provides a reliable and efficient alternative to experimental approaches.
• RPIFSE can aid in understanding biological processes involving RNA-protein interactions.