About

STAID is a deep learning framework for iterative deconvolution of spatial transcriptomics data.
It combines graph signal processing with pseudo-spot refinement to align single-cell references with spatial data, enabling accurate inference of cell-type compositions.

Key Features of STAID

  • Refined Pseudo-spot Generation
    Estimates likely cell types in each spot to create pseudo spots that better approximate real biological mixtures.

  • Graph Signal Processing with Fourier Transform
    Encodes expression profiles as graph signals, applies graph Fourier transform, and uses low-pass filtering to denoise, extracting robust features for accurate inference.

  • Iterative Pseudo-spot Refinement
    Generates new pseudo spots based on previous deconvolution results, progressively improving accuracy and robustness.

  • Reliable Cell-type Composition Estimation
    Provides accurate and stable deconvolution across diverse spatial transcriptomics datasets, enabling downstream studies of tissue architecture and cellular heterogeneity.

Applications

STAID provides a versatile framework for mapping cell-type distributions and gaining insights into tissue architecture and heterogeneity.