Skip to content

PSSUN/STMiner

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Static Badge Static Badge Static Badge Static Badge Static Badge Static Badge Documentation Status

Introduction

Why STMiner?

ST data presents challenges such as uneven cell density distribution, low sampling rates, and complex spatial structures. Traditional spot-based analysis strategies struggle to effectively address these issues. STMiner explores ST data by leveraging the spatial distribution of genes, thus avoiding the biases that these conditions can introduce into the results.


Method detail

Here we propose “STMiner”. The three key steps of analyzing ST data in STMiner are depicted.

(Left top) STMiner first utilizes Gaussian Mixture Models (GMMs) to represent the spatial distribution of each gene and the overall spatial distribution. (Left bottom) STMiner then identifies spatially variable genes by calculating the cost that transfers the overall spatial distribution to gene spatial distribution. Genes with high costs exhibit significant spatial variation, meaning their expression patterns differ considerably across different regions of the tissue. The distance array is built between SVGs in the same way, genes with similar spatial structures have a low cost to transport to each other, and vice versa. (Right) The distance array is embedded into a low-dimensional space by Multidimensional Scaling, allowing for clustering genes with similar spatial expression patterns into distinct functional gene sets and getting their spatial structure.

Quick start by example

Please visit STMiner Documents for installation and detail usage.

import package

from STMiner import SPFinder

Load data

You can download the demo dataset from GEO, or you can also download them from STMOMICS. STMiner can read spatial transcriptome data in various formats, such as gem, bmk, and h5ad (see STMiner Documents).
We recommend using the h5ad format, as it is currently the most widely used and supported by most algorithms and software in the spatial transcriptomics field.

sp = SPFinder()
file_path = 'Path/to/your/h5ad/file'
sp.read_h5ad(file=file_path, bin_size=1)
  • The parameter min_cells was used to filter genes that are too sparse to generate a reliable spatial distribution.
  • The parameter log1p was used to avoid extreme values affecting the results. For most open-source h5ad files, log1p has already been executed, so the default value here is False.
  • You can perform STMiner in your interested gene sets. Use parameter gene_list to input the gene list to STMiner. Then, STMiner will only calculate the given gene set of the dataset.

Find spatial high variable genes

sp.get_genes_csr_array(min_cells=500, log1p=False)
sp.spatial_high_variable_genes()

You can check the distance of each gene by:

sp.global_distance
Gene Distance z-score
geneA 9998 5.5
geneB 9994 5.4
... ... 5.3
geneC 8724 5.2

The first column is the gene name, and the second column is the difference between the spatial distribution of the gene and the background.
A larger difference indicates a more pronounced spatial pattern of the gene.

Preprocess and Fit GMM

sp.fit_pattern(n_comp=20, gene_list=list(sp.global_distance[:1000]['Gene']))

n_comp=20 means each GMM model has 20 components.

Build distance matrix & clustering

# This step calculates the distance between genes' spatial distributions.
sp.build_distance_array()
# Dimensionality reduction and clustering.
sp.cluster_gene(n_clusters=6, mds_components=20) 

Result & Visualization

The result is stored in genes_labels:

sp.genes_labels

The output looks like the following:

gene_id labels
0 Cldn5 2
1 Fyco1 2
2 Pmepa1 2
3 Arhgap5 0
4 Apc 5
.. ... ...
95 Cyp2a5 0
96 X5730403I07Rik 0
97 Ltbp2 2
98 Rbp4 4
99 Hist1h1e 4

Visualize the distance array:

import seaborn as sns
sns.clustermap(sp.genes_distance_array)

To visualize the patterns:

Note: A cutting border of the original dataset is needed to better visualize images. Anyhow, you can download the processed image here.

sp.get_pattern_array(vote_rate=0.3)
img_path = 'path/to/downloaded/image'
sp.plot.plot_pattern(vmax=99,
                     heatmap=False,
                     s=5,
                     reverse_y=True,
                     reverse_x=True,
                     image_path=img_path,
                     rotate_img=True,
                     k=4,
                     aspect=0.55)

Visualize the intersections between patterns 3 & 1:

sp.plot.plot_intersection(pattern_list=[0, 1],
                          image_path=img_path,
                          reverse_y=True,
                          reverse_x=True,
                          aspect=0.55,
                          s=20)

To visualize the gene expression by labels:

sp.plot.plot_genes(label=0, vmax=99)

Attributes of STMiner.SPFinder Object

Attributes Type Description
adata Anndata Anndata for loaded spatial data
patterns dict Spatial distributions pattern of genes
genes_patterns dict GMM model for each gene
global_distance pd. DataFrame Distances between genes and background
mds_features array embedding features of genes
genes_distance_array pd. DataFrame Distance between each GMM
genes_labels pd. DataFrame Gene name and their pattern labels
plot Object Call plot to visualization

Release history

https://pypi.org/project/STMiner/#history

Contact

Languages

  • Python 100.0%