• Python versions
  • Required packages
  • Tutorial
• MATLAB version
  • Installation
  • GUI Tutorial
  • Command-Line Tutorial
    • Preprocessing
    • Imputation
• Help

WEDGE

WEDGE is a biased low-rank matrix completion algorithm for recovering scRNA-seq gene expression data with high dropout rate.

Python versions

WEDGE's python version currently only supports single-core mode, and the multi-core version will be released soon. If you want to use the multi-core version or are familiar with MATLAB, we recommend that you use WEDGE's MATLAB version

Required packages

pip3 install numpy==1.19.3
pip3 install pandas
pip3 install scipy

Tutorial

For a detailed tutorial, please refer to test_WEDGE.ipynb

Attention:
Linux: WEDGE was compiled under the environment of Ubuntu20.04 and python3.8.6. If you want to use WEDGE in other environments, please recompile "NNLS.cpp" as follows:
    1. pip3 install pybind11
    2. Modify the path in the file "CMakeLists.txt" based on the path of pybind11
    3. mkdir build
    4. cd build
    5. cmake ..
    6. make
Windows: WEDGE was compiled under the environment of Windows10, vs2017 and python 3.8.6. If you want to use WEDGE in other environments, please recompile "NNLS.cpp" as follows:
    1. pip3 install pybind11
    2. Install Visual Studio(https://visualstudio.microsoft.com/zh-hans/vs)
    3. Modify the path in the file "setup.py"
    4. Execute "python .\setup.py build_ext --inplace" to recompile NNLS.cpp.

MATLAB version

Installation

If you are familiar with MATLAB, then please skip this section and view the Command-Line Tutorial. If not, we also provide GUI version below:

WEDGE Release	Windows	Linux desktop	Mac
Version_1.1	64-bit	64-bit	64-bit

• Windows & Mac: Double-click the installer to install WEDGE and select the default options for each step.

• Linux: If you have version 9.6 of the MATLAB Runtime installed in /mathworks/home/application/v96,run the shell script as:

     ./run_WEDGE.sh /mathworks/home/application/v96

  If you have MATLAB installed in /mathworks/devel/application/matlab, run the shell script as:

     ./run_WEDGE.sh /mathworks/devel/application/matlab-MCRInstaller.zip

Note: If you want to use WEDGE in Linux without desktop, please use MATLAB source code of WEDGE and view Command-Line Tutorial.

GUI Tutorial

Note: If using WEDGE to impute 10x data, one should select three files matrix.mtx , barcodes.tsv and genes.tsv at the same time, and upload them to WEDGE.

Command-Line Tutorial

Preprocessing

We recomend that user filter unwanted genes and cell before recovering raw data. By default, we normalize the total expression of each cell to 10,000, and perform log-transform after adding a pseudocount of 1.

Note: If there is a batch effect in the data, and the batch effect is not caused by dropout, it is recommended that the user remove the batch effect using other software such as seurat3.0.

Imputation

WEDGE_recovery() is the main function to recovery scRNA-seq data. We provided main_demo.m file to introduce how to change the parameters used in WEDGE. Here, You’ll get to know how to use WEDGE.

Initially, Begin by cleaning up the matlab environment:

clc;
clear;
close all;

Now, you should set the input file, a tsv or csv format file in which the first columns stores the gene names and the first row the cell names mandatorily；alternatively，the input file can be a 10x-Genomics-formatted mtx directory.

input_path= './Data/Baron.csv'; %the dir of input file;

Attention:

1. apostrophe '' is a must for input_path parameter due to internal implemention.
2. If input file is 10x-Genomics-formatted mtx directory, input_path must end with /, for example input_path= './Data/amount/'

Then, some optional paramaters can be set as below:

output_format =[]; %the saver format of recovery data.
%there are two options, 'W_H' and []
%if you set output_format = 'W_H', the output is W and H, the recovery data is equal to W*H;
%the default is output_format =[]; save recovery data as WEDGE_recovery.csv, rows is genes and columns is cells.

output_path =''; %the dir of output file, for example './res/'. 
%default output_path =[]; In this time if output_format = 'W_H', then output_path ='./res/'; otherwise, output_path ='./';

normalization = 1; 
%if normalization = 0, WEDGE do not normalize data.
%default, normalization = 1, WEDGE normalizes the total expression of each cell to 10,000 and performs log-transform after adding a pseudocount of 1. 


options.n_rank = []; 
%A positive integer, the rank of gene expression matrix; if options.n_rank = [], WEDGE will automatically estimate the rank of matrix
%otherwise, WEDGE will use the value set by the user.

n_CPU = []; 
%if n_CPU = [], WEDGE will use all the CPU in this computer, otherwise,
%WEDGE use min(n_CPU, the number of CPU in this computer) CPU to impute

options.lambda=[];
%the weights of  Zero elements;
%there are two select for  lambda, default lambda = 0.15 
%1. setting postive number to lambda

options.max_iters=[];% the maximum number of iterations for solving weight low-rank factorization;

options.tol=[]; %the error of object function when iter is convergence;

options.bound =[];%the minimum change value of single values when estimating the rank of gene expression matrix;

options.figure_state=0; %1 plot the change value of single values ; 0 don't plot the change value of single values . default 0;

Attention: output_path paramater must comply with the same rule as input_path illustrated above

Now, The last thing you have to do is to invoke the WEDGE_recovery() function.

[W,H]= WEDGE_recovery(input_path,output_path,output_format,normalization, n_CPU, options);

Help

If you have any questions or require assistance using WEDGE, please contact us: hyl2016@mail.ustc.edu.cn .