GxE models

G×E models

The one-environment model estimates additive genetic variance, interaction variance, residual variance, and environment-dependent residual variance. It can reuse one reference calculation across traits.

Inputs

Supply genotypes, a numeric environment table, covariates, and optional SNP annotations. Environment and covariate tables start with FID IID. The reference uses one fixed set of samples and variants.

1. Build the reference

summit --geno reference.bed --env environment.tsv --covar covariates.tsv \
  --nvecs 1024 --seed 1 --rand-dist rademacher --memory-gib 16 \
  --num-threads 8 --out results/reference

The default standardizes the environment and adjusts for the intercept, environment, and covariates. Genetic and interaction feature columns are normalized after that adjustment. The reference can be reused only with compatible SNPs, annotations, and feature definitions.

2. Compute trait summaries

summit --gxe-score-reference results/reference.gxe.ref.json \
  --geno reference.bed --env environment.tsv --covar covariates.tsv \
  --gxe-pheno phenotypes.tsv --gxe-pheno-cols trait1,trait2 \
  --num-threads 8 --out results/scores

All selected traits must be finite on the reference sample set. SUMMIT adjusts and normalizes the phenotypes and computes marginal additive and interaction scores in one genotype pass.

A conventional conditional interaction statistic, such as PLINK's ADDxE coefficient, is not the marginal interaction score required here. Use SUMMIT's trait summaries unless you have verified the generating formula independently.

3. Fit a trait

summit --gxe-fit results/reference.gxe.ref.json \
  --gxe-gwas results/scores.trait1.gxe.gwas.tsv.gz \
  --gwis results/scores.trait1.gxe.gwis.tsv.gz \
  --gxe-moments results/scores.trait1.gxe.moments.json \
  --njack 200 --out results/trait1

Results include .gxe.results.tsv, .gxe.fit.json, and .gxe.log. Standard errors use SNP-block deletion of the completed reference and trait summaries. Reference LD scores are held fixed during those deletions.

Use a new output prefix. --gxe-overwrite explicitly permits replacing an existing result.

Many traits or environments

--gxe-fit-batch fit_batch.json fits multiple traits against one reference. The manifest identifies each trait's moments, GWAS, GWIS, and output prefix:

{
  "kind": "summit.gxe.fit_batch",
  "schema_version": 1,
  "reference": "reference.gxe.ref.json",
  "traits": [{
    "name": "trait1",
    "moments": "scores.trait1.gxe.moments.json",
    "gwas": "scores.trait1.gxe.gwas.tsv.gz",
    "gwis": "scores.trait1.gxe.gwis.tsv.gz",
    "out": "fits/trait1"
  }]
}

--gxe-env-cols exposure1,exposure2 builds independent one-environment models while sharing genotype reads. The environments must have the same complete sample set. For a joint model that includes covariance between environmental responses, see Multiple environments.

A different study cohort

--gxe-population-reference allows scoring one selected trait in a different cohort. Supply that cohort's genotypes, environment, covariates, and phenotype, with --gxe-pheno-col. The model requires matching variant and feature definitions and assumes compatible genotype–environment distributions between reference and study. Cohort-specific residual moments are computed in the study.

Alternative scaling

--gxe-kernel-mode raw_projected --gxe-genotype-scale hwe uses the natural projected column norms of HWE-scaled genotypes for comparison with GENIE's kernel definition. It changes the variance-component interpretation. Keep reference and trait scaling consistent and report the chosen mode.

Methods describes the feature definitions and reference transfer.