Nucleotide-level prediction of CircRNA-protein binding based on fully convolutional neural network

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Circular RNAs (circRNAs) and protein binding are vital in biological processes and diseases.
• Current deep learning methods for circRNA-protein binding site prediction focus on sequence specificity but struggle with motif site accuracy.
• Accurate prediction of motif sites is essential for understanding their functional roles in gene expression.

Purpose of the Study:

• To develop a novel deep learning approach for accurately identifying circRNA-protein binding motif sites.
• To improve the prediction of functional motif sites involved in circRNA-protein interactions.
• To elucidate the role of these motifs in gene expression regulation.

Main Methods:

• Utilized fully convolutional neural networks (FCNNs) for nucleotide-level binary classification of circRNA-protein binding sites.
• Adapted computer vision pixel-level classification principles to sequence data.
• Employed the MEME tool and existing databases for motif analysis.

Main Results:

• Introduced CPBFCN, a new method for predicting circRNA motifs.
• Analyzed motif functions and their correlation with circRNA sponge activity.
• Identified potential contributions of flanking sequence motifs to circRNA-protein binding.

Conclusions:

• CPBFCN offers a new strategy for identifying circRNA-protein binding motifs.
• The findings enhance the understanding of circRNA-protein interactions in gene expression.
• This research aids in deciphering the regulatory roles of circRNA-protein binding.