Abstract <p><?tk 4?>Cluster analysis is essential in single-cell multi-omics research, allowing for simultaneous analysis across different omics levels. It reveals cellular diversity and precisely differentiates cell types and their physiological states. Despite advancements in clustering methods for single-cell multi-omics, challenges persist in analyzing the complexity of multi-omics data, integrating information across dimensions, ensuring dynamic adaptability, and maintaining consistent representation. Existing methods often fall short in capturing the full complexity and diversity of the data, tend to overemphasize local details while neglecting global patterns, and lack flexible clustering objectives. To tackle these limitations, we introduce scHEAGC, a novel hybrid encoding and adaptive guidance framework for enhanced single-cell multi-omics clustering. Our model employs a hybrid encoding strategy, integrating discrete and continuous techniques with vector quantization encoders (VQ-VAEs) and graph autoencoders (GAEs) to improve data representation. Furthermore, we design a multi-layer information fusion strategy that robustly integrates multi-dimensional data at both local and global levels, leveraging diverse embedding types. Additionally, our model includes an adaptive guided clustering module that dynamically adjusts targets in response to changes in the loss function, promoting iterative optimization. The hybrid encoding and the adaptive guided information fusion strategies we present are effective for single-cell multi-omics clustering and extendable to other fields. Experimental results demonstrate that scHEAGC outperforms existing methods across six real datasets, offering a powerful new tool for single-cell multi-omics data analysis.</p> Graphical Abstract <p></p>