The updated version of AWGAN with a new name ResPAN can be found here:https://github.com/AprilYuge/ResPAN. Now this paper has been accepted by Bioinformatics. The website of this model can be found here: https://respan.readthedocs.io/en/latest/.
Codes for paper: AWGAN: a powerful batch correction model for scRNA-seq data
We refer WGAN-GP and iMAP to construct the deep learning model and train it.
To install this tool, please use this code:
pip install awgan
To run our method, the first thing is to import necessary packages:
import awgan
import scprep
import numpy as np
import pandas as pd
import graphtools as gt
import os
import scanpy as sc
from skmisc.loess import loess
import torch.autograd
import torch.nn as nn
from torch.autograd import Variable
import torch.nn.functional as F
from collections import Counter
Then we need to load the scRNA-seq data with batch information:
adata = sc.read_loom('/content/drive/MyDrive/data/DC.loom', sparse=False) # use human dentritic dataset as one example
# scanpy.pp.highly_variable_genes` with `flavor='seurat_v3'` expects raw count data.
adata.X = np.float32(np.int32(adata.X))
adata = awgan.preprocessing.data_preprocess(adata) #preprocessing
Now we can assign the training sequence and generate required input information:
c = Counter(adata.obs['batch'])
c_keys = list(c.keys())
adata1 = adata[adata.obs['batch'] ==c_keys[0]]
adata2 = adata[adata.obs['batch'] !=c_keys[0]]
We can utilize this step to train the AWGAN, and get output_results as correction matrix:
output_results, model = awgan.model.sequencing_train(adata1,adata2,c_keys, epoch=40)
For more information and other examples, please take a look at 'ExampleCode' folder.
To run AWGAN and other benchmarks, we suggest you install the python environment with the version-specific package listed in this table.
| Package | Version | |
|---|---|---|
| Python packages | pytorch | 1.9.0 |
| scanpy | 1.8.1 | |
| scIB | 0.1.1 | |
| scprep | 1.1.0 | |
| umap-learn | 0.5.1 | |
| phate | 1.0.7 | |
| imap | 1.0.0 | |
| scVI | 0.6.8 | |
| bbknn | 1.5.1 | |
| mnnpy | 0.1.9.5 | |
| Harmonypy | 0.0.5 | |
| rpy2 | 3.4.5 | |
| R packages | kBET | 0.99.6 |
| LISI | 1.0.0 | |
| liger | 1.0.0 | |
| Seurat | 4.0.6 |
Please use:
AWGAN: a powerful batch correction model for scRNA-seq data. Tianyu Liu, Yuge Wang, Hong-yu Zhao. bioRxiv 2021.11.08.467781; doi: https://doi.org/10.1101/2021.11.08.467781