Sparser2Sparse: Single-shot Sparser-to-Sparse Learning for Spatial Transcriptomics Imputation with Natural Image...

Area of Science:

• Biomedical research
• Genomics
• Computational biology

Background:

• Spatial transcriptomics (ST) offers high-resolution gene expression profiling in tissues.
• High costs and data scarcity limit the widespread use of high-resolution ST.
• There is a need for cost-effective methods to reconstruct ST data.

Purpose of the Study:

• To develop a novel framework, Single-shot Sparser-to-Sparse (S2S-ST), for accurate spatial transcriptomics imputation.
• To enable robust ST reconstruction using only a single, low-cost, sparsely sampled ST dataset and natural images.
• To reduce the dependency on expensive high-resolution ST data.

Main Methods:

• Implemented a sparser-to-sparse self-supervised learning strategy to leverage intrinsic spatial patterns.
• Utilized cross-domain co-learning with natural images for enhanced feature representation.
• Employed a Cascaded Data Consistent Imputation Network (CDCIN) for iterative refinement and data fidelity preservation.

Main Results:

• S2S-ST demonstrated superior imputation accuracy compared to state-of-the-art methods across diverse tissue types (breast cancer, liver, lymphoid).
• The framework successfully reconstructs spatial transcriptomic data from sparse inputs.
• Achieved significant improvements in imputation accuracy, outperforming existing approaches.

Conclusions:

• S2S-ST provides a cost-effective solution for spatial transcriptomics imputation.
• The framework facilitates broader adoption of ST in biomedical research and clinical applications.
• Enables robust ST reconstruction from limited, low-cost data.