scPreGAN

Introduction

The official implementation of "scPreGAN, a deep generative model for predicting the response of single cell expression to perturbation". For reproducing the paper results, please visit XXX.

Installation

pip install git+https://github.com/XiajieWei/scPreGAN.git

Example

Example for out of sample prediction data of scRNA-seq data is shown in XXX. The data utilize in scPreGAN is in the h5ad format. If your data is in other format, please refer to Scanpy' or anndata's tutorial for how to create one.

Input

The data input to scPreGAN are better the normalized and scaled data, you can use follow codes for this purpose.

sc.pp.filter_genes(adata, min_counts=10)

sc.pp.filter_cells(adata, min_counts=3)

sc.pp.normalize_per_cell(adata)

sc.pp.log1p(adata)

sc.pp.scale(adata)

How to use it

from scPreGAN import *

# load data
train_data = load_anndata(path=data_path,
                condition_key=condition_key,
                condition=condition,
                cell_type_key=cell_type_key,
                out_of_sample_prediction=out_of_sample_prediction,
                prediction_cell_type=cell_type
                )

# create model
model = Model(n_features=n_features, n_classes=n_classes, use_cuda=True)

# training
model.train(train_data=train_data)

# predicting
pred_perturbed_adata = model.predict(control_adata=control_adata,
                   cell_type_key=cell_type_key,
                   condition_key=condition_key)