RSCNN-PseU: random searching-based convolutional neural network model for identifying RNA pseudouridine.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• RNA pseudouridine is a crucial post-transcriptional modification impacting RNA structure and function.
• Accurate identification of RNA pseudouridine is essential for understanding various biological processes.
• Existing methods for pseudouridine identification may lack efficiency or accuracy.

Purpose of the Study:

• To develop a more effective feature extraction method for identifying RNA pseudouridine.
• To propose a novel computational model for enhanced RNA pseudouridine detection.
• To improve the accuracy and adaptability of pseudouridine identification across different datasets.

Main Methods:

• RNA sequences were converted into numerical sequences using dinucleotide physicochemical properties (free energy, hydrophilicity).
• Discrete Fourier Transform (DFT) was applied to extract amplitude features, generating 2(N-1) features for a sequence of length N.
• A convolutional neural network (CNN) with a dynamic fully connected layer was employed for prediction, optimized using a random search algorithm.

Main Results:

• The proposed RSCNN-PseU model achieved a better identification effect for RNA pseudouridine.
• The model demonstrated effective feature extraction and accurate prediction capabilities.
• Adaptive regulation of model complexity was achieved, accommodating diverse species and datasets.

Conclusions:

• The novel feature extraction method combined with the RSCNN-PseU model offers a powerful approach for RNA pseudouridine identification.
• This method enhances the accuracy and efficiency of detecting this important RNA modification.
• The findings contribute to advancing research in RNA biology and related fields.