scsurv 0.2.0

a deep generative model for single-cell survival analysis

scSurv

Description

scSurv is a novel method for single-cell survival analysis. This model offers three main analyses:

1. Quantification of individual cells' contributions to clinical outcomes
2. Identification of prognosis-associated gene sets
3. Mapping of spatial hazard distributions using spatial transcriptome data

Through these applications, scSurv provides a comprehensive and interpretable framework that reveals heterogeneity in the clinical significance of cells. This framework enables the identification of novel cell populations and genes involved in the prognosis.

Teppei Shimamura's lab, Institute of Science Tokyo, Tokyo, Japan

Yasuhiro Kojima's lab, National Cancer Center Research Institute, Tokyo, Japan

The scSurv framework

1. The VAE compresses raw gene expression into low-dimensional latent cell representations
2. Deconvolution of bulk data
3. Estimation of regression coefficients in hazard functions for each cell using the extended Cox proportional hazards model

Installation

You can install scSurv using pip command.

!pip install scsurv

You can use the latest development version from GitHub.

!git clone https://github.com/3254c/scSurv.git

Dependencies

Python >= 3.8.16

torch >= 1.13.1

lifelines >= 0.27.8

scanpy >= 1.9.5

pandas >= 1.5.3

numpy >= 1.23.5

matplotlib >= 3.7.2

scipy >= 1.10.1

Minimal example

Preparing scRNA-seq data and bulk RNA-seq data

You need to prepare reference scRNA-seq data and bulk RNA-seq data linked to clinical outcome. For instance, TCGA bulk RNA-seq data is available from the Genomic Data Commons (GDC) Data Portal (https://portal.gdc.cancer.gov/). For examples of reference scRNA-seq data, please refer to our paper.

Run scSurv

In the tutorial, we demonstrate the usage of scSurv using simulated data. The bulk_adata.obs must contain the survival outcome data and the event indicators (censoring status). The scRNA-seq and bulk RNA-seq data must be raw counts. Please specify the batch_key parameter to indicate the sample origin annotation for each cell.

batch_key = 'orig.ident'
exp_name = 'tutolial_simulation.pt'
epoch = 10000

sc_adata, bulk_adata, model_params_dict, spatial_adata, scsurv_exp = workflow.run_scSurv(sc_adata, bulk_adata, exp_name, epoch, batch_key)