Fix GWAS fine-mapping postprocessing export

[Yining97]

The GWAS path through mnm_postprocessing had three defects that only show up on a real GWAS region with signal, so the toy/empty-region tests never hit them.

1. pip_sum crash on GWAS results

fine_mapping_cis_db_export.R builds the pip_sum table by reading db$context and db$trait. A GwasFineMappingResult has neither column (it's study, method, region_id, entry), so db$context is NULL and as.character(db$context[i]) is character(0) — length 0. On a region with signal getPip returns a real vector and the row hits data.frame(pip_sum = <len 1>, condition = character(0)), which errors with "arguments imply differing number of rows: 1, 0". Empty regions return early (getPip NULL), which is why it stayed hidden. The db and meta are written before this point, so only pip_sum is lost — and the step fails at the very end.

Fix: read each identity field NULL-safely, substituting NA when the column is absent. getPip already accepts context = NA.

2. export_top_loci silently empty under --gwas

Step 1 writes *.block_results_db.rds in GWAS mode, but export_top_loci_2 hardcoded export_suffix = 'cis_results_db.rds'. Under --gwas the glob matches nothing, stop_if skips gracefully, and step 3 combines over nothing and writes a 0-row .top_loci.bed.gz with exit 0 — a silent empty result unless you knew to pass --export_suffix block_results_db.rds by hand.

Fix: default export_suffix to empty and derive it from the gwas flag (block_results_db.rds under --gwas, else cis_results_db.rds); an explicit --export_suffix still overrides.

3. export_top_loci_3 combine aborts on macOS

The combine ran zcat "$f" without -f on its two emit lines; BSD zcat fails on .gz without -f, and set -euo pipefail aborts the step once fix #2 finally produces a non-empty bed to combine. Added -f, matching the zcat -f the same block already uses in its guard. No-op on GNU/Linux.

Validation

Real chr21 AD_Bellenguez FMRs, pecotmr 0.6.11 + susieR 0.16.6, full sos run.

• cped and replaced the workflow folder with the one in ADSPFG #1: baseline crash reproduced; patched pip_sum matches an independent sum(getTopLoci(rds)$pip[pip>0]) oracle on two GWAS objects (4.9979, 4.5563; condition=NA) and on a QTL result (9.8873; real context, unchanged).
• cped from old repo #2/clean up and add documentation #3: all four export_top_loci runs exit 0; combined bed is 24 rows for GWAS-default / QTL-default / explicit-override vs 0 rows for the unpatched baseline; combined variant_ID set matches getTopLoci(median05, 0.025) exactly.

No pecotmr change; no change to fine-mapping numbers — this is the export layer only.

🤖 Generated with Claude Code