sldsc_postprocessing_pipeline: align sd_annot names to polyfun .results categories (covers --snp-list mode)

[al4225]

Summary

sldsc_postprocessing_pipeline() matches sd_annot_full (named after .annot.gz columns) against polyfun's .results.Category. The previous code only worked when polyfun preserved the original annot column name. This PR makes the matching robust to all four pipeline configurations (single | joint × snp-list | no).

Background — the naming asymmetry

Polyfun appends _<file_idx> to LD score column names when writing .results.Category (target = file_idx=0). The LD score column name itself depends on which polyfun script wrote the file, and the new pipeline branches on --snp-list:

Branch	polyfun script	LD score col	.results target
no --snp-list, single	compute_ldscores.py	preserved ANNOT	ANNOT_0
--snp-list, single	ldsc.py --l2	hardcoded L2 (ldsc.py:317)	L2_0
no --snp-list, joint (N)	compute_ldscores.py	preserved A1, A2, …	A1_0, A2_0, …
--snp-list, joint (N)	ldsc.py --l2	<annot>L2 per col	A1L2_0, A2L2_0, …

compute_sldsc_annot_sd() only reads .annot.gz, so its return names never see the polyfun side. Without alignment, intersect() was empty whenever --snp-list was used → empty target_categories → cryptic downstream failure.

Fix

Two-stage match in sldsc_postprocessing_pipeline:

1. paste0("_0") on sd_annot_full / is_binary_full names (covers no-snp-list).
2. If intersect() is empty, take the first length(sd_annot_full) rows of .results.Category as targets and rename positionally (covers snp-list). Polyfun puts file_idx=0 rows first in .results, so position alignment is safe across all 4 branches.

Why this approach

• Pipeline-side renaming would reproduce ldsc.py's hardcoded L2/<annot>L2 rule — brittle.
• A use_snp_list flag from pipeline → pecotmr would couple the two repos.
• Trusting polyfun's row ordering (target before baseline because file_idx=0 < 1) needs no flag and works for any future polyfun script with the same convention.

Validation

• Standalone match test on real .results: ADSP allm (ANNOT_0, stage 1) and 1000G allm_snplist (L2_0, stage 2) both resolve correctly.
• MWE end-to-end (validate_pecotmr_fix.sh): COMPLETED 0:0 / 9:41 — stage 1 path unchanged.
• Production (1000G allm_snplist + m50_snplist, 6 contexts × 96 traits): all 6 produced <ctx>.sldsc_postprocess.rds (per_trait[96] + meta{tau_star, enrichment, enrichstat}) and 5 <group>.meta.rds per context — stage 2 confirmed at scale.

Test plan

• Reproduce MWE: bash xqtl-protocol/test/scripts/validate_pecotmr_fix.sh
• Run on a --snp-list .results (target = L2_0) and confirm fallback message in logs
• Run on a no-snp-list .results (target = ANNOT_0) and confirm no fallback
• Joint smoke: 2-col annot + --snp-list, verify target_categories = <a>L2_0, <b>L2_0

Files

• R/sldsc_wrapper.R (+32 lines, single function modified)

🤖 Generated with Claude Code