Compare tAge between experimental groups

Estimated marginal-mean contrasts of predicted tAge between a reference group and one or more comparison groups, as used for the clock analyses in the paper.

Usage

tage_compare_groups(
  data,
  value_columns,
  group_column,
  reference_group,
  compare_groups = NULL,
  covariates = NULL,
  split_by = NULL,
  se_columns = NULL,
  method = c("trt.vs.ctrl", "pairwise"),
  variance_strata = c("subset", "all_data"),
  p_adjust = TAGE_P_ADJUST_METHODS,
  p_adjust_scope = c("within_column", "across_columns", "global", "none"),
  conf_level = 0.95
)

Arguments

data

Data frame of per-sample predictions, typically the table returned by predict_tAge.

value_columns

Character vector of columns holding tAge predictions. Several columns are accepted so that p_adjust_scope = "across_columns" can correct across clocks or modules.

group_column

Column defining the experimental groups.

reference_group

Level of group_column used as the reference.

compare_groups

Levels compared against the reference. Default NULL uses every other level present.

covariates

Character vector of covariate columns entering the model as fixed effects. Default NULL.

split_by

Column to stratify on, fitting a separate model per level (e.g. "Tissue"). Default NULL.

se_columns

Columns holding per-sample prediction standard deviations from Bayesian ridge clocks. NULL (default) fits ordinary linear models. May be a single name, a vector parallel to value_columns, or a vector named by value column; NA entries fall back to lm.

method

"trt.vs.ctrl" (default) contrasts every comparison group against the reference; "pairwise" contrasts all pairs.

variance_strata

"subset" (default) estimates the residual variance from the compared groups only; "all_data" fits the model on every group present in the stratum and reports the requested contrasts from it.

p_adjust

Method passed to p.adjust; "BH" by default.

p_adjust_scope

Family over which p-values are corrected. "within_column" (default) corrects across all comparisons and strata of one clock. "across_columns" corrects across clocks within each comparison and stratum (the family used for module clocks). "global" corrects everything together, "none" disables it.

conf_level

Two-sided confidence level for ci_low / ci_high. The critical value follows the test: normal for the meta-regression, t otherwise.

Value

A data frame with one row per contrast and the columns value_column, split, group1 (reference), group2, n, estimate (always group2 - group1), se, statistic, df, p_value, p_adjusted and label.

Details

For elastic net clocks the model is value ~ group + covariates fitted with lm. For Bayesian ridge clocks – signalled by supplying se_columns – the model is a REML meta-regression (rma.uni) that weights each sample by its own prediction uncertainty, and the contrasts are z-tests rather than t-tests. In both cases contrasts come from emmeans with no built-in multiplicity adjustment; correction is applied afterwards over the scope given by p_adjust_scope.

See also

tage_regress_continuous for numeric predictors, tage_module_stats for module clocks, and tage_adjust_covariates for the matching plotting values.

Examples

if (FALSE) { # \dontrun{
results <- predict_tAge(tAge_eset, model_paths, species = "mouse", mode = "EN")
tage_compare_groups(
  results,
  value_columns   = "yugene_diff_EN_tAge",
  group_column    = "Genotype",
  reference_group = "WT",
  split_by        = "Tissue"
)
} # }