SpotGF

SpotGF: Denoising Spatially Resolved Transcriptomics Data Using an Optimal Transport-Based Gene Filtering Algorithm https://doi.org/10.1016/j.cels.2024.09.005

preprint paper:Optimal Transport Method-Based Gene Filter (GF) Denoising Algorithm for Enhancing Spatially Resolved Transcriptomics Data

GRAPHICAL ABSTRACT_revised

Installation

We recommend using conda to manage the installation of all dependencies. To do this, simply run:

conda create --name SpotGF
conda activate SpotGF

# conda config --add channels conda-forge ##If you do not have this channel added before#
conda install python=3.7 pandas pot=0.8.2 numpy scipy matplotlib descartes scanpy=1.9.2 seaborn

Then, download this repo and install it.

git clone [repo_path]
cd [path/to/SpotGF]
pip install .

The total installation time is around 5 mintunes. If error occuors, please upgrade pip and try again.

Usage

Once the input data have been processed into the supported format, the full SpotGF workflow can be run by calling the SpotGF.py script. The input files can include various formats such as gem, txt, csv, gem.gzand others, containing the raw SRT data information. These files must contain specific columns including geneID, x, y, MIDCount.

The denoising resolution can be adjusted using the binsize parameter. A smaller binsize will result in a finer denoising effect but will also increase the processing time.Please note that the binsize parameter is only used when calculating SpotGF scores and it does not affect the final denoised data.

If the input file contains columns named ‘cen_x’ and ‘cen_y’, the denoising process can be performed based on the cell bin data, achieving denoising at the single-cell level by setting ‘binsize = 1’ or ‘-b 1’.

The proportion parameter determines the proportion of genes to be retained in the final denoised data. For example, setting proportion=0.9 will retain only 90% of the effective genes, resulting in a new denoised SRT dataset.

#python [path/to/SpotGF.py] -i [path/to/input.gem] -b [binsize] -p [proportion] -s [spotsize]

cd /path/test

python ../SpotGF.py -i ./demo.gem -o ./result -b 70 -p 0.5 -s 5

If you use Spot in Jupyter environment, you can choose blow Usage.

import SpotGF   

gem_path = './demo.gem'  #please change this file path base on your work directory

binsize =70
lower= 0
upper = 100000
proportion  = 0.1
max_iterations = 10000
auto_threshold = True
visualize=True
spot_size=5
alpha=0

#initial class
spotgf =SpotGF.SpotGF(gem_path,binsize,proportion,auto_threshold,lower,upper,max_iterations,output,visualize,spot_size,alpha)

#acquire SpotGF socres distribution figure to help choose proportion 
GF_df = spotgf.calculate_GFscore(gem_path,binsize,alpha)

#denoised SRT data
new_gem  = spotgf.generate_GFgem(gem_path,GF_df,proportion,auto_threshold,visualize,spot_size)

Data preparation

SpotGF workflow requires two mandatory files, a input.gem (with "\t" or "," as the delimiter) for gene expression data,

geneID	x	y	MIDCount
#gene1	23	36	1
#gene1	24	35	1
#gene1	23	30	1
#gene2	20	31	1
#gene2	21	22	1

Output files

"SpotGF_scores.txt": This file contains the SpotGF scores for each gene. The SpotGF score indicates the degree of clustering or diffusion of a gene. A smaller SpotGF score suggests that the gene is more diffuse, while a larger SpotGF score indicates that the gene is more clustered.

"SpotGF_proportion_0.5.gem": A gene expression matrix file with noise reduction that retains only 50% of the genes.

"SpotGF_auto_threshold.gem": Gene expression matrix file after denoising based on an automatic filtering threshold method.

"alpha_shape.png": It shows the tissue contour used by SpotGF during the denoising process, with special attention to the fact that a highly detailed outline is not required; only a rough outline is needed.

"Spatial_automatic.png": Spatial expression levels of data after denoising based on an automatic filtering threshold method.

"Spatial_proportion.png": Spatial expression levels of data after denoising based on a manual threshold method.

"Violinplot_n_genes_by_counts.png": Violin plot distribution of n_genes_by_counts across several datasets.

"Violinplot_total_counts.png": Violin plot distribution of total_counts across several datasets.

List of Parameters

positional arguments:

  -i                Input SRT data files path.

  -o                Outpath for saving results.

  -b                Denoising resolution binsize, must int type, default=10.

  -lower            Lower limit for tissue structures capturing optimization, default=0.

  -upper            Upper limit for tissue structures capturing optimization', default=sys.float_info.max.

  -max_iterations   maximum number of iterations when capturing tissue structures', default=10000.

  -p                Proportion of gene numbers, must float type [0,1], default=0.5.

  -auto_threshold   if True, return a denoised gem file using automatic threshold, default=True.

  -v                Visualize SpotGF-denoised data, default=True.

  -s                Spot size used for spatial expression figure, default=5.

  -a                Alpha for tissue boundary detection, default use auto optimizealpha, default=0.

Contact Us

If you have any suggestions/ideas for SpotGF or are having issues trying to use it, please don't hesitate to reach out to us.

Lin Du, dulin@genomics.cn

Cite

Du, L., Kang, J., Hou, Y., Sun, H. X., & Zhang, B. (2024). SpotGF: Denoising spatially resolved transcriptomics data using an optimal transport-based gene filtering algorithm. Cell systems, 15(10), 969–981.e6. https://doi.org/10.1016/j.cels.2024.09.005

illuminate6060 / SpotGF

readme