Dataset for SC2CD

Single-cell spatial transcriptomics enables the study of cellular organization. However, computational techniques for spatial clustering and cell-type decomposition are often required, as spot-based platforms such as Visium lack single-cell resolution. Although these two tasks are inherently interconnected, current approaches typically address them separately, limiting information exchange and reducing overall accuracy. We introduce SC2CD, an iterative framework that co-optimizes spatial clustering and cell decomposition through bidirectional information sharing. SC2CD integrates deep learning-based spatial clustering with matrix factorization-based decomposition, dynamically refining spatial graphs using decomposition-informed similarities while leveraging clustering results to improve local deconvolution. Unlike conventional methods that treat these tasks in isolation, SC2CD supports iterative refinement, enabling mutual enhancement. SC2CD consistently outperforms existing approaches, producing more precise spatial clusters and higher-resolution cell-type identification. By unifying spatial clustering and cell-type deconvolution within a co-optimization framework, SC2CD improves spatial transcriptomic resolution and facilitates a more comprehensive understanding of complex tissue architecture. The datasets included here were used to evaluate and validate the SC2CD framework described in our study.