Spot2vector

Spot2vector is a novel computational framework that leverages a ZINB-based graph autoencoder for spatial clustering and data denoising. This method integrates both spatial and expression information to provide a comprehensive analysis of spatial transcriptomics (ST) data.

Authors

• [email protected]
• [email protected]

Pipeline

Requirements

1. Anaconda or Miniconda: Ensure you have either Anaconda or Miniconda installed.
2. CUDA version >= 11.8: Required for GPU acceleration.
3. NVIDIA GPU available: Ensure you have a compatible NVIDIA GPU.

Installation

For detailed installation instructions, please refer to INSTALLATION.md

Quick Start

1. Data Preparation

The input data for Spot2vector should be an AnnData object, which can be loaded using scanpy.read_h5ad. The AnnData object must contain:

• Preprocessed Expression Data: The expression data should be preprocessed using standard single-cell RNA-seq preprocessing steps:

  import scanpy as sc
  
  # Normalize total counts
  sc.pp.normalize_total(adata, target_sum=1e4)
  
  # Log transform the data
  sc.pp.log1p(adata)
  
  # Select highly variable genes
  sc.pp.highly_variable_genes(adata, n_top_genes=8000, flavor='seurat_v3')

• Spatial coordinates: The spatial coordinates should be stored in adata.obsm["spatial"]. The coordinates should be a 2D array of shape (n_spots, 2).

• Optional PCA: For improved efficiency in constructing the expression similarity graph, you can perform PCA to obtain a low-dimensional representation:

  sc.pp.pca(adata, n_comps=10)

2. Graph Construction

Construct spatial and expression graphs using the following commands:

import spot2vector

# Spatial graph based on spatial coordinates
spot2vector.Build_Graph(adata, radius_cutoff=150, cutoff_type='radius', graph_type='spatial')

# Expression graph based on expression similarity
spot2vector.Build_Graph(adata, neighbors_cutoff=4, cutoff_type='neighbors', graph_type='expression')

3. Model Training

Train the model using the following command:

device = 'cuda:0'  # Specify the GPU device
spot2vector.Fit(adata, device=device)

4. Spatial Clustering (Spatial & Expression)

Perform spatial clustering using both the expression embeddings and spatial embeddings. The n_clusters parameter specifies the number of spatial domains, and users need to provide this value based on their dataset and biological knowledge.

# Expression embeddings
spot2vector.Clustering(adata, obsm_data='exp_embeddings', method='mclust', n_cluster=n_clusters, verbose=False)

# Spatial embeddings
spot2vector.Clustering(adata, obsm_data='spa_embeddings', method='mclust', n_cluster=n_clusters, verbose=False)

5. Model Inference

Perform model inference to obtain the final embeddings:

# lamda = 1 for expression, lamda = 0 for spatial
spot2vector.Infer(adata, lamda=0.2, device=device)

6. Spatial Clustering (Final Embeddings)

Perform the final spatial clustering using the combined embeddings:

spot2vector.Clustering(adata, obsm_data='embeddings', method = 'mclust', n_cluster=n_clusters, verbose=False)

License

This project is licensed under the MIT License - see the LICENSE file for details.