Batch analyses

Batch analyses

Batch modes reduce repeated input loading when many traits share an LD-score and annotation model.

Many heritability estimates

summit --h2 sumstats/ --ldscores reference.gw.ldscore.gz \
  --annot annotations.tsv --h2-batch-fast --njack chr \
  --h2-batch-size 4 --h2-workers 2 --num-threads 4 --out results/h2

Fast h² supports HE weighting and chromosome jackknife. For integer SNP blocks, LDSC weighting, or chi-square clipping, use regular --h2.

For repeated annotation models, add --h2-cache-dir cache/h2. The default read/write cache can be reused when the SNP order is identical. After building it, --h2-cache-mode read prevents new entries. --h2-cache-only prepares or checks entries without fitting.

Each cached trait needs about 8.125 bytes per SNP for its values and active mask, before metadata. Loader workers, batch size, and BLAS threads control different resources; increasing all three can increase memory substantially.

Many genetic correlations

A tab-delimited manifest specifies one pair per row:

phen1 phen2 sumstats1 sumstats2 overlap_covariance cov_rank1 cov_rank2
trait_a trait_b trait_a.sumstats trait_b.sumstats 0 0 0

summit --rg pairs.tsv --ldscores reference.gw.ldscore.gz \
  --rg-manifest-fast --njack chr --out results/rg

Fast rg requires a supplied finite overlap covariance in every row, HE weighting, and jackknife standard errors. Ordinary manifests can mix supplied and summary-estimated overlap covariance. Use regular mode for other weighting or uncertainty methods.

Outputs include batch.log, manifest.results.tsv, and per-pair results. --rg-fast-no-pair-logs keeps only the batch log and combined table.

Chromosome jackknife reproduces regular supplied-overlap fits. Integer block mode uses blocks defined before pair-specific filtering and may differ from regular mode's block assignment.

Many annotation models

Store all unique annotation columns and corresponding LD scores once. A model manifest selects each model's columns:

model bins aliases
model_a ["base","annotation_a"] ["base","focal"]
model_b ["base","annotation_b"] ["base","focal"]

Use tabs between fields and the same ordered aliases for each model.

summit --rg pairs.tsv --rg-manifest-fast --rg-model-manifest models.tsv \
  --ldscores union.gw.ldscore.gz --annot union.annot.tsv \
  --njack chr --out results/models

SUMMIT loads shared traits and SNP statistics once, then fits each selected model. Verify that shared columns are identical before combining existing annotation datasets.

PGS batching is described in Polygenic scores.

Cross-trait environmental responses

CrossTraitBatch estimates the ordered context-score products for multiple trait pairs in one genotype traversal. Each trait retains its own samples and fixed effects. Within-trait and cross-trait fits use aligned SNP blocks so their uncertainty can be propagated jointly.

See Cross-trait analysis for the Python workflow, reference assumptions, and interpretation of response correlations.