CIRI-Deep Enables Single-Cell and Spatial Transcriptomic Analysis of Circular RNAs with Deep Learning

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Circular RNAs (circRNAs) exhibit tissue- and cell-type-specific expression.
• Current single-cell and spatial transcriptomics struggle to profile circRNAs effectively due to sequencing limitations.

Purpose of the Study:

• To develop a deep learning model for comprehensive circRNA regulation prediction.
• To overcome limitations in profiling circRNAs from diverse RNA-seq data.

Main Methods:

• Development and training of the CIRI-deep model on 25 million circRNA regulation events.
• Validation of CIRI-deep performance on test and leave-out datasets.
• Application of CIRI-deep for circRNA detection, visualization, and feature evaluation.

Main Results:

• CIRI-deep demonstrates high accuracy in predicting circRNA regulation from RNA-seq data.
• The model successfully infers differential circRNA events.
• CIRI-deep enables various analyses, including region-specific detection and feature importance.

Conclusions:

• CIRI-deep provides a robust method for circRNA analysis across diverse RNA-seq data types.
• The model's adaptability extends to single-cell and spatial transcriptomics.
• CIRI-deep is poised to significantly advance circRNA research.