Commands and options

Commands and options

Use summit --help for analysis options, or add --help to a command such as summit pgs fit.

Commands

Task	Command
Genome-wide or windowed LD scores	summit --geno ...
Heritability or genetic correlation	summit --h2 ... or summit --rg ...
Binary-trait preparation and inference	summit --binary-method pcgc ...
One-environment G×E	summit --geno ... --env ..., --gxe-score-reference, --gxe-fit, or --gxe-fit-batch
Polygenic scores	summit pgs {plan,fit,score,scale,inspect}
Generalized G×E reference planning and inspection	summit reference {plan,inspect}
Cross-trait reference Z moments	summit reference zpass

Generalized G×E fitting and cross-trait response models use Python APIs. See Multiple environments and Cross-trait analysis for their workflows.

Shared controls

Option	Meaning
--geno	BED or PGEN input, with companion files
--annot	SNP annotations
--out	Output prefix or directory, as specified by the command
--nvecs	Number of random vectors for reference estimation
--seed	Random seed
--block-size	Number of genotype variants processed together
--num-threads	Compute thread count
--memory-gib	Working-memory budget in GiB
--njack	Jackknife block count or scheme

Use complete option names. --block-size controls computation; --njack controls uncertainty estimation. Windowed LD scores do not use random vectors. --block-size auto is available for one-environment G×E reference generation. Its resolved width is saved with the reference because it affects the finite-vector estimate.

Workflow	Random vectors	Seed	Block size	Memory budget
LD/G×E preparation	1,000	Unspecified	1,000	auto
Binary preparation	256	0	256	1 GiB
PGS	—	—	512	16 GiB
Generalized reference plan	Required	—	4,096	Required
Reference Z moments	—	—	128	—

Binary inference uses 200 contiguous SNP blocks by default. HE/LDSC uses chromosome deletion (--njack chr). Input formats and method-specific settings are described in the individual analysis guides.

Memory

For LD/G×E, --memory-gib budgets the sketch panels or windowed-LD workspace. For binary preparation it budgets reference workspace; for PGS it supplies the fit or scoring memory plan. It is not a limit on process resident memory. Allow additional memory for inputs, libraries, and output arrays.

G×E also provides --gxe-native-workspace-gib for each direct native call (default 16 GiB), and --gxe-total-memory-gib for the estimated process peak (default auto). --win-cache-mb controls the windowed-LD cache in MiB. These limits cover different allocations.

Genome build

Provide build-matched genotypes, annotations, and reference files. SUMMIT does not convert coordinates or infer the genome build.

--genome-build optionally records a label during binary or PGS scale preparation. The PGS specification accepts the equivalent genome_build field. Labeled saved scales must be reused with the same label. PGS scoring rejects conflicting labels when both are supplied and always checks SNP positions and allele pairs, including when labels are omitted.