scAce: an adaptive embedding and clustering method for single-cell gene expression data
Project description
scAce: an adaptive embedding and clustering method
Overview
scAce is consisted of three major steps, a pre-training step based on a variational autoencoder, a cluster initialization step to obtain initial cluster labels, and an adaptive cluster merging step to iteratively update cluster labels and cell embeddings. In the pre-training step, scAce takes the single-cell gene expression matrix as its input to train a VAE network. For each gene, the VAE learns and outputs three parameters of a ZINB distribution (mean, dispersion, and proportion of zero). In the cluster initialization step, scAce offeres two manners. With de novo initialization, Leiden is used to obtain initial cluster labels; with clustering enhancement, initial cluster labels are obtained by applying a cluster splitting approach to a set of existing clustering results. In the adaptive cluster merging step, given the pre-trained VAE network and the initial cluster labels, the network parameters, cell embeddings, cluster labels and centroids are iteratively updated by alternately performing network update and cluster merging steps. The final results of cell embeddings and cluster labels are output by scAce after the iteration process stops.
Installation
You can install scAce
from pypi with:
pip install scace
Or clone this repository and using
pip install -e .
in the root of this repository.
Usage
Load the data to be analyzed:
import scanpy as sc
adata = sc.AnnData(data)
Perform data pre-processing:
# Basic filtering
sc.pp.filter_genes(adata, min_cells=3)
sc.pp.filter_cells(adata, min_genes=200)
adata.raw = adata.copy()
# Total-count normlize, logarithmize and scale the data
sc.pp.normalize_per_cell(adata)
adata.obs['scale_factor'] = adata.obs.n_counts / adata.obs.n_counts.median()
sc.pp.log1p(adata)
sc.pp.scale(adata)
Run scAce algorithm:
from scace import run_scace
adata = run_scace(adata)
The output adata contains cluster labels in adata.obs['scace_cluster']
and the resulting embedding in adata.obsm['scace_emb']
, this embedding can be used as input of other downstream task.
To see an example, please refer to tutorial.ipynb
.
Reproduce results
See folder reproducibility
.
Follow README.md
in data
and pkgs
to prepare data and packages.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.