Covariate-adjusted tAge values for plotting — tage_adjust

Removes the fitted covariate effects from a tAge column while keeping the group effect, and adds the mean covariate effect back so the adjusted values stay on the original scale.

Usage

tage_adjust_covariates(
  data,
  value_column,
  covariates,
  split_by = NULL,
  se_column = NULL,
  group_column = NULL
)

Arguments

data: Data frame of per-sample predictions.
value_column: Column to adjust.
covariates: Covariate columns to regress out.
split_by: Optional stratifying column. With group_column, one model per stratum, as in tage_compare_groups; without it, a fixed effect for elastic net clocks and one model per stratum for Bayesian ridge clocks.
se_column: Per-sample standard deviations for a Bayesian ridge clock. Default NULL uses lm.
group_column: Column holding the experimental groups, fitted together with the covariates. Default NULL keeps the covariate-only model.

Value

Numeric vector of adjusted values, in the row order of data; rows with missing values in the model return NA.

Details

With group_column the covariate effects are those of the model the statistics fit – value ~ group + covariates, one model per stratum, lm or the weighted meta-regression – so the difference between group means of the adjusted values is the tested estimate (exactly for elastic net clocks with variance_strata = "all_data"). Without it the covariate model leaves the group out; when the covariates are unevenly distributed across groups (more females among the treated, say) that model attributes part of the group effect to the covariates, and the adjusted values no longer show what was tested.

Examples

if (FALSE) { # \dontrun{
results$adjusted <- tage_adjust_covariates(
  results, "yugene_diff_EN_tAge", covariates = "Sex", split_by = "Tissue",
  group_column = "Genotype"
)
} # }