Empowering Multi-faceted Analysis of Spatial Transcriptomics Data with RGAST

RGAST: Reproducibility Scripts This repository contains the custom scripts and notebooks used to reproduce the analysis and figures presented in the paper "Empowering Multi-faceted Analysis of Spatial Transcriptomics Data with RGAST". The code is organized by analysis task. Below is a detailed description of the directory structure and the purpose of each file. Directory Structure 1. spatial_clustering Core scripts for spatial domain identification, including benchmarking and robustness tests. DLPFC_RGAST.py: Main script for running RGAST clustering on 12 DLPFC slices. MERFISH.py / Mouse_Stereo.py: RGAST clustering scripts for MERFISH and Stereo-seq datasets. seqFISH_RGAST.ipynb: Notebook for RGAST clustering on the seqFISH+ dataset. sensitivity_analysis.py: Script to perform hyperparameter sensitivity analysis (varying neighbor counts and ). struct_noise_analysis.py: Script for the robustness analysis (random node dropping from 0% to 50%). sensitivyty.R: R script to visualize the hyperparameter sensitivity heatmap. DLPFC.ARI_boxplot.R / SC_boxplot.R: R scripts to plot ARI and Silhouette Coefficient comparisons. 2. CCC_analysis (Cell-Cell Communication) Scripts for inferring cell-cell interactions (CCI) and benchmarking communication networks. CCI_HDST.ipynb: Notebook for analyzing cell-cell interactions on High-Definition Spatial Transcriptomics (HDST) data. CCI_MERFISH.ipynb: Notebook for analyzing cell-cell interactions on MERFISH data. *.h5ad: Processed AnnData files used as input for the notebooks (BC_sub.h5ad, animal1_0.26.h5ad). simulation/: Files related to the simulation study for benchmarking CCI inference accuracy. simulate_gt_spatial.py: Generates synthetic spatial data with ground-truth interaction networks. benchmark_simu.R: R script to compare inferred networks against ground truth. corr.R: Calculates correlation metrics for interaction strengths. sensitivity_plot.R: R script to visualize the sensitivity analysis results. data/: Contains simulated ground truth matrices (truth_adj_*.mtx) and RGAST inferred matrices (RGAST_adj_*.mtx) for validation. 3. SVG_detection (Spatially Variable Genes) Scripts for benchmarking SVG detection methods and analyzing spatial patterns. DLPFC.py / MERFISH.py: Scripts to run RGAST's SVG detection module on DLPFC and MERFISH datasets, respectively. SpatialDE_DLPFC.py: Script to run the baseline method SpatialDE on DLPFC slices (added during revision). run_spatialde.sh: Shell script to batch process SpatialDE on multiple samples. svg_utils.py: Helper functions for calculating Moran’s I and Geary’s C statistics. SVG_cluster.ipynb: Notebook for clustering analysis using identified SVGs as features. seqFISH.ipynb: Notebook for visualizing SVGs on the seqFISH+ dataset. SVG_boxplot.R: R script to plot the Moran's I/Geary's C comparison boxplots (Fig. 4). 4. trajectory Scripts for trajectory inference and pseudotime analysis. save_emb_umap.py: Extracts latent embeddings from the trained RGAST model and saves them for downstream trajectory analysis. monocle3_analysis.R: R script to perform Monocle3 trajectory inference using RGAST embeddings, including pseudotime calculation and correlation analysis with cortical layers. 5. 3D_reconstruction Scripts for 3D spatial domain identification and quantitative evaluation of cross-slice continuity. 3D_RGAST.ipynb: Main notebook for training the RGAST model on multiple aligned tissue slices to generate 3D spatial domains. cross_slice_consistency.py: Python script to calculate quantitative metrics (Label Transfer Accuracy and Embedding Cosine Similarity) for evaluating the continuity of 3D reconstruction across adjacent slices. Getting Started Prerequisites Please refer to the root requirements.txt (if available) or the main RGAST repository for the environment setup. Key dependencies include: Python >= 3.8 (PyTorch, Scanpy, NumPy, Pandas) R >= 4.0 (Monocle3, ggplot2, Seurat)