scE2EGAE

Enhancing Single-cell RNA-Seq Data Analysis through an End-to-End Cell-Graph-Learnable Graph Autoencoder with Differentiable Edge Sampling

The preprint manuscript at Research Square

Introduction about code scripts and hyperparameters

Codes

distances.py: functions used for distances metrix calculation

modules.py: codes of all the modules we used. The modules.py mainly contains the following classes:

           a) EdgeSamplingGumbel(nn.Module): For edge sampling 
           
           b) GAE(torch.nn.Module): The graph autoencoder for single-cell RNAseq data denoising 
           
           c) ZINBAE(Module): The DCA module for projecting single-cell features (counts) into a lower-dimensional space.

losses.py: codes of the loss functions

imputation_model.py: the intergrated model comprises from the modules

train.py: the pytorch train wrapper codes

Main.ipynb: codes to train the model and some simple visualization

Hyperparameters: please follow this table when initializing the model

Datasets	Num_Epochs	Patience	K	Distance_Measure	AE_Dim	GAE_Dim	Dropout_GAE	LR	Alpha	Beta	MSE_V2
Klein	500	20	3	Hyperbolic	128	2000	0	0.003	0.0005	1	False
Zeisel	700	30	3	Hyperbolic	128	2000	0.1	0.003	0.001	1	True
Romanov	500	20	3	Hyperbolic	128	2000	0.1	0.003	0.0005	1	False
ITC	500	20	1	Hyperbolic	128	2000	0.1	0.003	0.001	1	False
Chu	2000	20	1	Euclidean	128	2000	0	0.003	0.001	1	False
ILC	4000	10	1	Hyperbolic	128	2000	0	0.003	0.001	1	False
Tirosh	800	10	1	Hyperbolic	128	128	0	0.003	0.001	1	False
AD	1500	20	1	Hyperbolic	128	64	0	0.003	0.001	1	False

How to run the code

1. First, we create an Anaconda environment:

conda create -n pytorch python=3.9

conda activate pytorch

2. Then, install the needed packages:

conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.7 -c pytorch -c nvidia

conda install pyg -c pyg

pip install anndata==0.8.0

pip install notebook

pip install h5py==3.7.0

pip install loompy==3.0.7

pip install matplotlib==3.7.1

pip install numpy==1.23.4

pip install pandas==1.5.1

pip install scanpy==1.9.3

pip install scikit-learn==1.1.3

pip install scipy==1.9.3

3. Clone the repo to your PC:

git clone https://github.com/PPDPQ/scE2EGAE.git

4. If you have raw data, you can use the notebooks in the /data_prep folder. If you do not have them, due to GitHub's file upload size limit of 25MB, we have uploaded a processed Zeisel dataset to the "/dataset" folder. You can try running our code using this dataset.

5. Run the "Main.ipynb" notebook cell by cell.