cell-decipher 0.1.0

Spatial-omics data embedding and analysis

DECIPHER

Installation • Documentation • Citation • FAQ • Acknowledgement

DECIPHER aims to learn cells’ disentangled embeddings from intracellular state and extracellular contexts view based on single-cell spatial omics data.

Installation

[!IMPORTANT] Requires Python >= 3.10 and CUDA-enabled GPU (CPU-only device is not recommended).

PyPI

We recommend to install cell-decipher to a new conda environment with RAPIDS dependencies.

mamba create -n decipher -c conda-forge -c rapidsai -c nvidia python=3.11 rapids=24.04 cuda-version=11.8 cudnn cutensor cusparselt -y && conda activate decipher
pip install cell-decipher
install_pyg_dependencies

Docker

Build docker image from Dockerfile or pull the latest image from Docker Hub by:

docker pull huhansan666666/decipher:latest

Documentation

Please check documentation for detailed tutorial.

Minimal example

Here is a minimal example for quick start:

import scanpy as sc
from decipher import DECIPHER

# Init model
model = DECIPHER(work_dir='/path/to/work_dir')

# Register data
adata = sc.read_h5ad('/path/to/adata.h5ad')
model.register_data(adata)

# Fit model
model.fit_omics()

# Get disentangled omics and spatial embeddings
omics_emb = model.center_emb
spatial_emb = model.nbr_emb

Demo

Name	Description
Basic Model Tutorial (Colab)	Tutorial on how to train DECIPHER
Identify Localization Related Genes	Tutorial on how to identify cells’ localization related genes via DECIPHER embeddings
Multi-slices with Batch Effects	Tutorial on how to remove batch effects across multiple slices
DDP Training	Tutorial on how to use multi-GPUs on large datasets

Citation

TBD

If you want to repeat our benchmarks and case studies, please check the benchmark and experiments folder.

FAQ

1. CUDA out of memory error

We do all experiments on A100-80G GPUs. We observed model.train_gene_select() may use ~40GB for 700,000 cells with 1,000 genes. If your GPU do not have enough memory, you can try running with CPU.

Acknowledgement

We thank following great open-source projects for their help or inspiration:

• SimCLR
• vit-pytorch
• lightly
• scib
• rapids_singlecell