Skip to content

Command-line interface¤

The jaxqtl executable provides four subcommands:

Command Result
jaxqtl compute-pcs Expression principal components appended to covariates
jaxqtl cis One lead association and adjusted p-value per tested phenotype
jaxqtl nominal Every association within each cis window
jaxqtl trans Chunked phenotype-by-variant associations

Run jaxqtl COMMAND --help for the complete parser-generated option list and defaults.

Common mapping options¤

All mapping commands require one genotype source plus --pheno and --covar.

Group Options
Genotypes --bfile, --pfile, --vcf, --bgen, --dosage
Covariates --covar-name, --rm-covar, --normalize-covar, --one-hot, --no-intercept
Library-size adjustment (offsets) --offset, --offset-name-from-covar, --set-offset-from-libsize
Model and variant testing --model, --test, --robust-se, --spa
Gene-level testing --acat, --nperm
Filters --keep, --exclude, --min-indiv-expr-pct, --min-gene-expr-pct, --maf, --chr
Phenotypes --gene-list, --genes, --window, --tss-centered
Solver --max-iter, --tol, --gtol, --step-size, --solver
Runtime --seed, --platform, --verbose, --out

Some accepted flags apply only to particular combinations. --robust-se requires a Wald test; --spa applies to score tests; --acat and --nperm affect only cis; and --window and --tss-centered affect only cis and nominal.

For score-test ACAT scans, --spa --acat is strongly recommended because ACAT is sensitive to variant p-value calibration. See Tests and gene-level calibration for why Beta permutation does not have the same dependence on asymptotic tail probabilities.

Mapping automatically retains expression phenotypes on chromosome labels shared with the genotype input. --chr further restricts both phenotypes and genotype variants to one exact label, which must occur in both inputs.

See the workflow guides for complete commands with compatible options.

GLM fitting controls¤

Option Default Meaning
--max-iter 1000 Maximum IRLS iterations
--tol 1e-3 Absolute change in total negative log likelihood that triggers the gradient check
--gtol 1e-3 Per-observation gradient tolerance, with coefficient scaling and NB2 bound projection
--step-size 1.0 Initial trial step for each IRLS update; rejected trials are halved
--solver cholesky Weighted least-squares solver; choices are cholesky, qr, and cg

Both likelihood and gradient criteria must be met for GLM convergence. These controls govern model fitting; they do not set the SPA root-solver or Beta-calibration tolerances. See Troubleshooting for interpretation and troubleshooting.