nf-fgwas

Nextflow implementation of fGWAS for single cell data. The pipeline uses the hierarchical model from here and the application to single cell data has been described in detail here.

In brief, given a single cell dataset and GWAS summary statistics, fGWAS allows to find cell types that are enriched for the genetic associations. It requires full GWAS summary statistics, including logOR and standard error.

run

# load modules (or provide nextflow and singularity differently)
module load cellgen/nextflow  # internal module with singularity-ce version 4.1.0
module load cellgen/singularity  # internal module with nextflow version 24.10.2

# make a new directory for the results
mkdir make/new/directory
cd make/new/directory

# run the pipeline
nextflow run /path/to/nf-fgwas/main.nf -resume \
    --tss_file "/path/to/tss_file.txt.gz" \
    --cell_types "/path/to/tss_file.txt" \
    --atac_file "/path/to/atac_file.bed.gz" \
    --broad_fine_mapping "/path/to/broad_fine_mapping.tsv" \
    --studies "/path/to/studies.csv"

Alternatively, copy and edit nextflow.config, then execute

nextflow -C nextflow.config run /path/to/nf-fgwas/main.nf

Note: if you want to use summary statistics stored on iRODS (Sanger internal only), run this before starting the pipeline and enter your password:

module load cellgen/irods
iinit

input files

GWAS studies

A list of GWAS studies is supplied via a studies.csv file. It should contain IDs and paths to multiple summary statistics files, one study per row.

Each study needs an ID (to name results folders) and summary statistics can be supplied in one of three ways:

1. path to a custom .bed.gz file
2. path to a .parquet file
3. only the study ID, in this case iRODS will be used to fetch a corresponding file called <studyID>.parquet from the Sanger farm (Sanger internal only)

The studies.csv file could look like this (Empty rows and lines starting with "#" will be ignored):

# fetch these studies from the farm
STUDY1
STUDY2

# parquet files
STUDY3,/path/to/study_3.parquet

# custom bed files
STUDY4,/path/to/study_4.bed.gz

custom files

Full summary statistics can often be downloaded in CSV/TSV format, e.g. from the EBI GWAS catalog in harmonised format. Make sure that they:

• are full summary statistics including SNP position, beta value, standard error
• have genome build version compatible with the transcription start sites in tss_cell_type_exp.txt.gz (e.g. GRCh38)

Further, the files need to be transformed into BED format with these columns:

• 1st column: chromosome number (e.g. 1)
• 2nd column: SNP position (e.g. 345435)
• 3rd column: SNP position (e.g. 345435, same as 2nd column)
• 4th column: other allele (e.g. A)
• 5th column: effect allele (e.g. G)
• 6th column: beta value (e.g. 0.5)
• 7th column: standard error (e.g. 0.5)

Finally, the BED file needs to be compressed in block format using bgzip and indexed using tabix. The index file <file_name>.bed.gz.tbi is expected to be in the same folder as <file_name>.bed.gz.

Note: a script is available in bin/check_summ_stat.py to help prepare custom files. See bin/check_summ_stat.py --help for more details. Often summary statistics provided for different studies do not contain all required values or the columns are labelled incorrectly. The script contains some checks to test the required values are present, as well as potentially converts them is other values are present that allow calculating the ones that are required (beta and standard error).

RNA data

Average RNA counts per cell type need to be supplied as a TSV file.

Note: a script is available in bin/prepare_input.py to help prepare this file from an AnnData object. See bin/prepare_input.py tss --help for more details.

The TSV file should be:

• a file called tss_cell_type_exp.txt.gz (tab separated and gzipped) with one row per gene and:
  • 1st column: chromosome number
  • 2nd column: chromosome position
  • next columns, one for each cell type: mean expression of gene in cell type
  • last column: overall mean expression for the gene
  • sorted ascending by first two columns
  • no header

(Currently, in addition to the gzipped file, the unzipped version including a header needs to be supplied (--cell_types). This is only to provide the cell type names and will be simplified in the future.)

ATAC data (optional)

Optionally for including ATAC data another BED file needs to be supplied.

Note: a script is available in bin/prepare_input.py to help prepare this file from an AnnData object. See bin/prepare_input.py atac --help for more details.

• a file ending .bed.gz (bgzip compressed and tabix indexed) that contains
  • first column: chromosome position
  • second column: peak start position
  • third column: peak end position
  • remaining columns, one for each cell type:
    • 1 or 0 depending on whether peak present in cell type or not
• a file with a cell type mapping between RNA and ATAC (--broad_fine_mapping)

The file provided to --broad_fine_mapping might look like this:

atac_cell_type    rna_cell_type
immune    B_cell
immune    T_cell
epithelial    Goblet
epithelial    Club

(note: cell type annotations can also be identical, e.g. if RNA/ATAC comes from the same cells)