Fine Mapping Workflow

BayesTME can take advantage of companion scRNA data.

Instead of using cross validation on the ST dataset to determine the cell types, cell types can be determined from the companion scRNA directly, using an established workflow such as Seurat.

The resulting relative expression values of each gene in each cell type/cluster (represented by φ_kg in equation 4 in the BayesTME preprint) can be provided to the deconvolution step via the --expression-truth option (CLI) or expression_truth= parameter in bayestme.deconvolution.deconvolve().

We have provided a docker container and script for running Seurat on 10Xgenomics/cellranger output, which will produce the appropriate relative expression values in a CSV output that can be read into the BayesTME pipeline. Example usage:

docker run \
    -v /path/to/local/cellranger/dir:/opt/data \ # path to raw_feature_bc_matrix directory in the cellranger output
    -v /path/to/local/output/dir:/out/output \ # output dir for CSV and plot results
    /jeffquinnmsk/bayestme-seurat-fine-mapping:latest \
    --dataDir /opt/data \
    --outputDir /opt/output

The output CSV, at <output dir>/average_expression.csv will be provided as the argument to deconvolve --expression-truth.

If you have multiple matched scRNA samples for a single ST dataset, you can run this workflow on all of them, and then provide the --expression-truth argument multiple times to deconvolve.

For example:

deconvolve --expression-truth average_expression_sample_1.csv \
    --expression-truth average_expression_sample_2.csv \
    --expression-truth average_expression_sample_3.csv

deconvolve will consider all the samples jointly to determine baseline expression profiles for the different cell types.