Skip to contents

Predict transcriptomic age with several clocks

Source: R/predict.R
predict_tAge.Rd

Applies clocks to the preprocessed representations returned by tAge_preprocessing and returns the sample metadata with one prediction column per clock.

Usage

predict_tAge(
  tAge_eset,
  model_paths,
  species = NULL,
  mode = NULL,
  return_std = NULL,
  age_units = c("auto", "months", "years"),
  normalized_age = c("fraction", "percent")
)

Arguments

tAge_eset

A named list of ExpressionSet objects, each representing a different normalization method (e.g., "scaled", "scaled_diff", "yugene", "yugene_diff"), as returned by tAge_preprocessing.

model_paths

A clock table with a path column, or a named list of model paths per representation; see Details.

species

Species of the samples: "mouse", "rat", "human" or "monkey" (see tage_species). Its only effect is the rescaling of chronological-age clocks to age units by the species maximum lifespan; mortality and normalized-age clocks ignore it. It is not the species group the model was trained on ("Mouse" / "Rodents" / "Multispecies" in list_clocks). Default NULL takes the species recorded by tAge_preprocessing.

mode

"EN" or "BR". Default NULL takes each model's type from the clock table, the registry or its file name.

return_std

Logical. Whether to keep the per-sample predictive standard deviation of Bayesian Ridge clocks. Default NULL keeps it for every Bayesian ridge clock, as a <column>_sd column. Pass these to the se_columns argument of tage_compare_groups for the Bayesian ridge statistics.

age_units

Units of chronological-age predictions: "auto" (default; months for rodents, years for primates), "months" or "years".

normalized_age

Scale of normalized-age clocks: "fraction" of the expected maximum lifespan (default) or "percent" (x100, as in the paper).

Value

A data frame containing the predicted transcriptomic age results for all provided ExpressionSet objects, with appropriately named columns. The attribute "tage_units" is a named character vector giving the unit of every prediction column (e.g. "months", "log10 hazard ratio").

Details

model_paths is either

  • a clock table from list_clocks or list_module_clocks with a path column (e.g. from download_clocks): each row is applied to the representation its scaling names (Scaled -> scaled_diff, YuGene -> yugene_diff), and the column is named after the model file, as in the Python package; or

  • a named list, representation -> model path(s). With one path per representation the column is <representation>_<mode>_tAge; with several, each column is named after its model file.

Examples

if (FALSE) { # \dontrun{
clocks <- download_clocks(list_clocks(type = "EN", species = "Rodents",
                                      tissue = "Multi-Tissue"))
res <- predict_tAge(tAge_eset, clocks)       # six columns, named by model file
attr(res, "tage_units")
} # }