Low-coverage imputation WDL pipeline

Short WDL workflows to impute low-coverage sequencing data with GLIMPSE2.

Contents

• bcf_imputation.wdl — Runs imputation for a single chromosome by scattering over chunk definitions; tasks: ImputeChunk (GLIMPSE2_phase) and Ligate (GLIMPSE2_ligate); outputs a per-chromosome BCF (.bcf + .bcf.csi) and logs/timings.
• launch_imputation.wdl — Master workflow that launches the single-chromosome workflow across a set of chromosomes and aggregates the per-chromosome outputs.
• *.json — Example inputs files for single-chromosome and whole-genome runs.

Environment

• Validated on whole-genome runs in Google Cloud using the Cromwell workflow engine.
• Reference panel and other inputs can be provided as gs:// paths.
• Tasks automatically fetch a Google Cloud OAuth token from the instance metadata server (GCS_OAUTH_TOKEN) so tools can stream from GCS without pre-downloading files.

Docker requirements

The runtime image must contain:

• GLIMPSE2 (GLIMPSE2_phase, GLIMPSE2_ligate, compiled inside the docker)
• htslib built with libcurl support (for authenticated HTTP/GCS streaming)
• bcftools (for indexing/concat, as needed)
• curl (metadata + remote access), jq (parse OAuth token), time (GNU time for timing)
• Standard GNU utilities used in commands: coreutils, grep, sed, awk, etc.

Example usage (Cromwell)

Single chromosome:

• Inputs file: bcf_imputation.json
• Run: java -jar cromwell.jar run bcf_imputation.wdl --inputs bcf_imputation.json

Whole genome:

• Inputs file: launch_imputation.json
• Run: java -jar cromwell.jar run launch_imputation.wdl --inputs launch_imputation.json

Outputs are written as <output_prefix>_chr<chr>.bcf (+ .bcf.csi) with accompanying log and timing files.