Toward simultaneous pseudo-space reconstruction and cell-type deconvolution of single-cell spatial transcriptome...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution cellular data but lacks spatial context.
• Spatial transcriptomics (ST) retains tissue spatial information but offers lower cellular resolution.
• Existing methods struggle to integrate these complementary modalities effectively.

Purpose of the Study:

• To introduce SpaDicer, a deep learning framework unifying pseudo-spatial reconstruction of single cells and cell-type deconvolution of ST spots.
• To bridge the gap between high-resolution scRNA-seq and spatially resolved ST data.
• To enhance the understanding of tissue heterogeneity through integrated spatial and cellular information.

Main Methods:

• Developed an end-to-end deep learning framework, SpaDicer.
• Employed cross-domain feature disentanglement for domain-invariant feature extraction.
• Utilized an adaptive loss weighting strategy to harmonize multiple learning objectives.

Main Results:

• SpaDicer demonstrated improved reconstruction of cellular spatial localization across diverse human tissue datasets.
• The framework showed enhanced cell-type deconvolution accuracy in ST data.
• Consistent performance improvements were observed compared to state-of-the-art methods on simulated and real-world data.

Conclusions:

• SpaDicer effectively unifies scRNA-seq and ST data, overcoming individual modality limitations.
• The multi-task learning framework advances spatially informed biological discovery.
• SpaDicer offers a powerful tool for detailed analysis of tissue architecture and cellular composition.