Skip to content

preprocessing.md

Latest commit

 

History

History
97 lines (62 loc) · 1.91 KB

preprocessing.md

File metadata and controls

97 lines (62 loc) · 1.91 KB

🧬 HEIST Preprocessing Guide

This script performs preprocessing on raw spatial transcriptomics data, constructing cell-level and gene regulatory networks (GRNs) and saving them in PyTorch Geometric (PyG) format for downstream tasks.

📂 Input Requirements

  • Raw .h5ad files should be placed in the directory:

    data/sea_raw/

  • Example file structure:

    data/sea_raw/sample1.h5ad
data/sea_raw/sample2.h5ad

📦 Dependencies

conda activate HEIST  # Or your preferred environment
pip install scanpy magic-impute torch-geometric scikit-learn tqdm networkx

🚀 How to Run

python utils/preprocess.py  # Replace with the actual filename

⚙️ What This Script Does

  1. Loads Raw Data: Loads .h5ad files and extracts metadata like Braak stages and spatial locations.

  2. Preprocesses Data:

    • Filters genes expressed in fewer than 3 cells.
    • Normalizes and log-transforms expression values.
    • Applies MAGIC for denoising.

  3. Constructs Graphs:

    • Cell-Level Graph: Based on spatial proximity using sc.pp.neighbors().

    • Gene Regulatory Networks (GRNs):

      • Built using Mutual Information (MI) between gene pairs.
      • Edges are thresholded at MI > 0.35.
      • Converted to PyG format using from_networkx().

  4. Saves Preprocessed Graphs:

    • Output is saved to:

      data/sea_preprocessed/

    • Each file contains a list of graphs:

      • One high-level cell graph.
      • Multiple low-level gene graphs (one per cell).

📁 Generated Output Example

  • Cell and gene graphs are saved as .pt files:

    data/sea_preprocessed/sample1_0.pt

📈 Expected Console Output

Loaded file sample1.
Now pre-processing the data.
Creating the GRNs using MI
Converting to PyG format
Data saved
Number of graphs in the dataset: 3500