Decoding protein binding landscape on circular RNAs with base-resolution transformer models.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Circular RNAs (circRNAs) are crucial regulators of gene expression.
• Existing methods for predicting RNA-binding protein (RBP) binding sites on circRNAs lack precision and resolution.
• Current tools often analyze circRNA fragments, failing to pinpoint specific binding sites on the full transcript.

Purpose of the Study:

• To develop a novel computational tool, CircSite, for precise prediction of RBP binding sites on circRNAs.
• To achieve single-nucleotide resolution in identifying RBP interactions with circRNAs.
• To pinpoint key nucleotides crucial for RBP binding on circRNA transcripts.

Main Methods:

• A hybrid deep learning approach combining Convolutional Neural Networks (CNNs) and Transformer models.
• Utilizing CNNs for local feature extraction and Transformer for global sequence representation.
• Training and benchmarking on 37 curated datasets of circRNA-protein interactions.

Main Results:

• CircSite accurately predicts RBP binding sites at single-nucleotide resolution on circRNA transcripts.
• The tool successfully identifies key nucleotides contributing to RBP binding.
• Predicted key subsequences align well with known RBP binding motifs, validating the model's efficacy.

Conclusions:

• CircSite offers a significant advancement in understanding RBP interactions with circRNAs.
• The tool provides a user-friendly webserver for researchers to predict RBP binding sites on circRNAs.
• CircSite facilitates deeper insights into circRNA regulatory mechanisms and functions.