Abstract
The rapid advancements in spatially resolved transcriptomics (SRT) enable the characterization of gene expressions while preserving spatial information. However, high dropout rates and noise hinder accurate spatial domain identification for understanding tissue architecture. We present DeepGFT, a method that simultaneously models spot-wise and gene-wise relationships by integrating deep learning with graph Fourier transform for spatial domain identification. Benchmarking results demonstrate the superiority of DeepGFT over existing methods. DeepGFT detects tumor substructures with immune-related differences in human breast cancer, identifies the complex germinal centers accurately in human lymph node, and accurately reveals the developmental changes in 3D Drosophila data.