sldsc: fix compute_sldsc_M_ref reference panel; add target_labels relabeling

[al4225]

Summary

Two changes to the sLDSC post-processing in R/sldsc_wrapper.R:

1. compute_sldsc_M_ref — count the reference panel, not the regression set

M_ref is the reference-panel SNP count over which partitioned heritability is defined:
h²(C) = Σ_{j∈M_ref} a_C(j)·Σ_{C'} τ_{C'}·a_{C'}(j). It is panel-defined (it matches polyfun's .l2.M for allm / .l2.M_5_50 for m50), not the regression / HapMap3 SNP set (~1M).

Previously, under maf_cutoff == 0 the function counted the .l2.ldscore rows of target_anno_dir. For snplist runs that directory holds the HM3 regression set (~1M), not the reference panel (~8M) — so allm + snplist runs reported a ~8× too-small M_ref, and therefore a ~8× too-small τ* / EnrichStat (e.g. meta τ* 0.018 instead of 0.137).

Now it counts the reference panel from the .frq files:

• maf_cutoff == 0 → all rows (matches polyfun .l2.M)
• maf_cutoff > 0 → MAF > cutoff rows (matches polyfun .l2.M_5_50)
• target_anno_dir is kept only as a fallback (with a warning) when no .frq dir is supplied.

Enrichment is unaffected (polyfun computes it independently of M_ref); only τ* / EnrichStat standardization changes. m50 / m50_snplist runs were already correct.

2. sldsc_postprocessing_pipeline — new optional target_labels argument

When given (same length & order as the resolved target_categories), every target column and tau*-block column name in the output is renamed to those labels; params$target_categories then holds the labels and params$target_categories_orig keeps the original polyfun .results names (ANNOT_0, L2_0, …). When NULL (default) nothing is renamed — original behaviour. Lets downstream output read e.g. quantile_eQTL instead of ANNOT_0.

Testing

End-to-end on the updated_pipeline_by_gao/test MWE (postprocess + meta_subset), both with and without --target-categories-label:

• with label → params$target_categories = quantile_eQTL, target_categories_orig = ANNOT_0, all target columns relabeled, meta_subset inherits the label;
• without label → unchanged (ANNOT_0, no target_categories_orig);
• numeric values (enrichment, τ*) identical between the two — relabel is cosmetic only.
• M_ref fix verified on ADSP_allm_snplist: M_ref 1.04M → 8.13M, meta τ* 0.018 → 0.137, Enrichment unchanged.

man/compute_sldsc_M_ref.Rd and man/sldsc_postprocessing_pipeline.Rd regenerated via roxygen2.

🤖 Generated with Claude Code