Analysis of Parkinson’s Disease Spatial Transcriptomics Dataset

Research Paper

📄 parkinsons_spatial_transcriptomics_research_paper.pdf

Data

GSE253975_data/

Raw data directory containing the downloaded files from
GEO: GSE253975.

ckpt/

Folder containing pretrained CellPLM checkpoint weights:
CellPLM Checkpoints (Dropbox).

adata_preprocessed.h5ad

Preprocessed AnnData object saved after running the preprocessing notebook (loading/merging counts, filtering, and CP10K + log-normalization).

Code

preprocessing.ipynb

Loads the raw Visium .txt.gz count matrices in GSE253975_data, inspects them, builds a combined AnnData with sample_id/batch metadata, filters zero-count genes, normalizes to CP10K + log1p, and saves the reusable adata_preprocessed.h5ad (with a small matrix preview at the end).

parkinsons_analysis.ipynb

Starts from adata_preprocessed.h5ad, selects 2k HVGs, scales, runs PCA, corrects batch effects with Harmony, builds neighbor graph + UMAP, and clusters with K-Means. It computes silhouette/compactness/batch-mixing metrics for PCA vs Harmony.

parkinsons_analysis_cellplm.ipynb

Also loads adata_preprocessed.h5ad but replaces PCA with 512-dim CellPLM embeddings (CPU fallback). Uses those embeddings for neighbors, UMAP, K-Means clustering, and silhouette/compactness/batch-mixing metrics.

How to Run

1. Install the required Python packages listed in requirements.txt.
2. Make sure the GSE253975_data/ directory and the ckpt/ checkpoint folder are in place.
3. Run preprocessing.ipynb to load the GEO files and generate the unified normalized adata_preprocessed.h5ad.
4. Run parkinsons_analysis.ipynb and parkinsons_analysis_cellplm.ipynb to perform their respective analyses and automatically generate figures in the figures/ folder.