Disease-state identification with healthy single-cell references

This repository contains notebooks and scripts to reproduce analyses benchmarking the use of control and atlas datasets as references for identification of disease-associated cell states (see manuscript).

The workflow for disease-state identification and evaluation of out-of-reference detection is available as a python package.

Repository structure

• diff2atlas - utility module
• metadata - metadata tables used for analysis
• src - analysis notebooks and scripts
  • 1_PBMC_data_preprocessing/ - preprocessing and harmonization of PBMC dataset
  • 2_simulation_design/ - out-of-reference detection benchmark on simulations
  • 3_simulation_ctrl_atlas_size - out-of-reference detection robustness to atlas and control dataset size
  • 3b_crosstissue_atlas - out-of-reference detection robustness with tissue-matched or cross-tissue atlas
  • 4_COVID_design - reference design comparison on COVID-19 dataset
  • 5_IPF_HLCA_design - reference design comparison on IPF lung dataset

Data

Processed datasets and scVI models used in this analysis are available via figshare. For references of the original datasets collected see study metadata.

For simulation analysis

• PBMC_merged.normal.subsample500cells.clean_celltypes.h5ad - harmonized object of healthy PBMC profiles from 13 studies, used for OOR identification benchmark with simulations
• model_PBMC_merged.normal.subsample500cells.zip - scVI model trained on healthy PBMC profiles (used for joint annotation) (trained with scvi-tools v0.16.2, see notebooks for training parameters)
• Results from simulation analysis are shared in .csv files (OOR_simulations_*.csv)
  • *.nhood_results_all.csv - neighbourhood level Milo results (with fraction of OOR state)
  • *.TPRFPRFDR_results_all.csv - TPR/FDR/FPR for each simulation
  • *.AUPRC_results_all.csv - AUPRC for each simulation

For COVID-19 analysis

• PBMC_COVID.subsample500cells.atlas.h5ad - atlas dataset (PBMCs from healthy individuals from 12 studies)
• PBMC_COVID.subsample500cells.covid.h5ad- disease dataset (PBMCs from COVID-19 patients from Stephenson et al. 2021)
• PBMC_COVID.subsample500cells.ctrl.h5ad - control dataset (PBMCs from healthy individuals from Stephenson et al. 2021)
• PBMC_COVID.subsample500cells.design.query_PC_refA.post_milo.h5ad - ACR design processed object with Milo results
• PBMC_COVID.subsample500cells.design.query_PC_refA.post_milo.nhood_adata.h5ad - ACR design processed object with Milo results (nhood AnnData)
• PBMC_COVID.subsample500cells.design.query_P_refC.post_milo.h5ad - CR design processed object with Milo results
• PBMC_COVID.subsample500cells.design.query_P_refC.post_milo.nhood_adata.h5ad - CR design processed object with Milo results (nhood AnnData)
• model_COVID19_reference_atlas_scvi0.16.2.zip - scVI model trained on atlas dataset (used for ACR design) (trained with scvi-tools v0.16.2, see script for training parameters)

For IPF analysis

• IPF_HLCA.ACR_design.post_milo.h5ad - ACR design processed object with Milo results. Includes annotation of aberrant basal-like states (adata.obs['basal_like_annotation'])
• IPF_HLCA.ACR_design.post_milo.nhood_adata.h5ad - ACR design processed object with Milo results (nhood AnnData)
• IPF_HLCA.CR_design.post_milo.h5ad - CR design processed object with Milo results.
• IPF_HLCA.CR_design.post_milo.nhood_adata.h5ad - CR design processed object with Milo results (nhood AnnData)
• IPF_HLCA.AR_design.post_milo.h5ad - AR design processed object with Milo results.
• IPF_HLCA.AR_design.post_milo.nhood_adata.h5ad - AR design processed object with Milo results (nhood AnnData)

For cross-tissue atlas analysis

• model_TabulaSapiens_scvi0.20.0.zip - scVI model trained on Tabula Sapiens dataset (trained with scvi-tools v0.20.0, see script for training parameters)

Citation

Dann E., Teichmann S.A. and Marioni J.C. Precise identification of cell states altered in disease with healthy single-cell references. biorXiv https://doi.org/10.1101/2022.11.10.515939

Contact

For any questions, please post an issue.