Predicting circRNA subcellular localization by fusing circRNA sequence and network information.

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Circular RNAs (circRNAs) are increasingly recognized for their crucial roles in biological processes.
• Understanding circRNA function necessitates determining their subcellular localization.
• Traditional experimental methods for localization are costly and time-consuming.

Purpose of the Study:

• To develop an efficient computational model, CircLoc, for predicting circRNA subcellular localization.
• To leverage circRNA sequences and network information for accurate prediction.

Main Methods:

• Feature extraction using k-mer, large language model (RNAErnie), and network representation learning (node2vec, graph attention auto-encoder).
• Integration of extracted features using a self-attention layer and a fully connected prediction layer.
• Model evaluation via ten-fold cross-validation.

Main Results:

• CircLoc achieved an average AUC of 0.7856 and AUPR of 0.4055.
• Performance surpassed traditional multi-label classification and miRNA localization prediction models.
• Ablation tests confirmed the model's effectiveness and feature importance.

Conclusions:

• CircLoc provides an effective computational approach for predicting circRNA subcellular localization.
• The model can serve as a valuable tool for advancing circRNA research.