CIRI-Deep enables single-cell and spatial transcriptomic analysis of circular RNAs (circRNAs) using deep learning. The model, trained on 25 million high-confidence circRNA regulation events, accurately predicts differential splicing events across diverse RNA-seq data types, including single-cell and spatial transcriptomic datasets. CIRI-Deep and its adapted version, CIRI-DeepA, can detect circRNA clusters or regions, visualize BSJ ratio maps, and evaluate trans and cis feature importance. The model's adaptability extends to all major RNA-seq datasets, broadening circRNA research. CIRI-Deep uses cis and trans features to predict circRNA regulation, achieving high accuracy in both test and leave-out data. It performs well on low-depth and poly(A)-selected data, enabling analysis of circRNAs in samples with limited sequencing depth. The model also identifies tissue-specific cis and trans features, revealing the role of RBPs and splicing factors in circRNA regulation. CIRI-DeepA is effective in predicting DSC events from poly(A)-selected RNA-seq data, including single-cell and spatial transcriptomic data. It outperforms traditional methods in detecting circRNAs in low-depth data and provides insights into circRNA expression in different tissues. CIRI-Deep and CIRI-DeepA demonstrate high accuracy in predicting circRNA splicing events, enabling the analysis of circRNA junction ratios in spatial transcriptomic data. The model's ability to infer cell type proportions from circRNA data highlights its potential in single-cell transcriptomic analysis. Overall, CIRI-Deep represents a promising solution for circRNA research, offering new insights into circRNA regulation and function.CIRI-Deep enables single-cell and spatial transcriptomic analysis of circular RNAs (circRNAs) using deep learning. The model, trained on 25 million high-confidence circRNA regulation events, accurately predicts differential splicing events across diverse RNA-seq data types, including single-cell and spatial transcriptomic datasets. CIRI-Deep and its adapted version, CIRI-DeepA, can detect circRNA clusters or regions, visualize BSJ ratio maps, and evaluate trans and cis feature importance. The model's adaptability extends to all major RNA-seq datasets, broadening circRNA research. CIRI-Deep uses cis and trans features to predict circRNA regulation, achieving high accuracy in both test and leave-out data. It performs well on low-depth and poly(A)-selected data, enabling analysis of circRNAs in samples with limited sequencing depth. The model also identifies tissue-specific cis and trans features, revealing the role of RBPs and splicing factors in circRNA regulation. CIRI-DeepA is effective in predicting DSC events from poly(A)-selected RNA-seq data, including single-cell and spatial transcriptomic data. It outperforms traditional methods in detecting circRNAs in low-depth data and provides insights into circRNA expression in different tissues. CIRI-Deep and CIRI-DeepA demonstrate high accuracy in predicting circRNA splicing events, enabling the analysis of circRNA junction ratios in spatial transcriptomic data. The model's ability to infer cell type proportions from circRNA data highlights its potential in single-cell transcriptomic analysis. Overall, CIRI-Deep represents a promising solution for circRNA research, offering new insights into circRNA regulation and function.