Linking microbial genetic signals and Single-Cell Transcriptomics to Unravel Host-microbe Interactions across Human Tissues

Introduction

In this study, we introduce a computational framework (scBPS) to incorporate microbial GWAS summary data with human single-cell transcriptomic data for the discovery of critical cellular contexts that are highly associated with gut microbial taxon.

Documentation

1, GWAS summary data are preprocessed to generate z-scores for microbe-gene associations. We rank the relevance of genes based on z-scores and choose top 1,000 genes for each taxon as microbe-associated genes, analogue to previous studies.

2, we collapse the expression of the putative microbe-associated genes in a given cell to calculate a raw Bacteria Polygenic Score (BPS) for each cell; to enhance the statistical power, these putative microbe-associated genes are weighted by corresponding MAGMA-estimated z-scores and inversely weighted by gene-specific technical noise scores, which are estimated by the mean-variance relationship across genes. The raw BPSs are subsequently subjected to gene set-wise and cell-wise standardization for obtaining a normalized BPS, which measures the strength of association between each taxon and each host cell.

3, to distinguish taxon-relevant cell types or tissue types, we estimate the relevance of each cell population using a BPS_AUC metric, which is determined by calculating the area under the recovery curve (AUC) of cells in the predefined population across the ranking of all cells’ BPSs in the single-cell atlas 35. To assess statistical significance, scBPS generates 1,000 sets of control BPS_AUC by randomly permuting the rank of all the cells in the single-cell atlas. Monte Carlo (MC) P value for each “taxon-cell population” pair is computed based on the empirical distribution of the BPS_AUC across the corresponding 1,000 control BPS_AUC (see Methods).

4, the output of our framework includes: (1) the BPS values of each individual host cell for each microbial taxon; (2) BPS_AUC scores representing the strength of microbe-cell type associations and the empirical P values.

Data analysis

• Preprocess of microbial GWAS summary data:

bash <myfile>/main-codes/run_removeMHC.sh <GWAS_file> <myfile>

• GWAS data SNP annotation:

mkdir <outfile1>
bash <myfile>/main-codes/annotation.sh <GWAS_file> <outfile1> <magmadir>

• gene z-scores:

mkdir <outfile2>
bash <myfile>/main-codes/gene_anno.sh <GWAS_file> <outfile1> <magmadir> <myfile> <outfile2>

# <scfile> is the output file.

• Calculating BPS scores:

mkdir <outfile3>
python <myfile>/main-codes/compute.score.py <outfile3> <outfile2>/zscore_merged.gs <single.cell.file>/humanatlas.h5ad

• Calculating BPS_AUC scores:

# cell type level

mkdir <outfile4>

R <myfile>/main-codes/calculate_BPSAUC_celltype.R  <outfile3> <outfile4> <myfile>

# tissue level

mkdir <outfile4>

R <myfile>/main-codes/calculate_BPSAUC_tissue.R  <outfile3> <outfile4> <myfile>

• Monte-Carlo p-values:

# cell type level

mkdir <outfile5>

R <myfile>/main-codes/MonteCarlo_celltype.R <outfile4> <outfile5> <myfile>

# tissue level

mkdir <outfile5>

R <myfile>/main-codes/MonteCarlo_tissue.R <outfile4> <outfile5> <myfile>

Resources

The microbial GWAS summary data of the Dutch microbiome project were downloaded at https://dutchmicrobiomeproject.molgeniscloud.org.

The Tabula Sapiens human single-cell transcriptome data were downloaded at https://tabula-sapiens-portal.ds.czbiohub.org/.

The GWAS summary data of MiBioGen project were downloaded at https://www.mibiogen.org/.

The GWAS summary data of 10 liver-associated diseases were downloaded from the FinnGen database at https://r8.risteys.finngen.fi/.

All codes used for the analyses and visualization are provided in the Github repository at https://github.com/jjlea/scMicrobiome.

References