plot_contribution_to_sample: Plot feature contributions to sample predictions

In Mushriq/mixmap: Test package to share mixmap code

Plot feature contributions to sample predictions

Description

This function plots the contribution (Shapley values) of the top features to a sample's predictions.

Usage

plot_contribution_to_sample(
  model,
  model_data,
  name,
  sample_names,
  n_features = 5,
  n_columns = 1,
  short_title = FALSE,
  fixed_axis = FALSE,
  axis_limits = c(-0.05, 1),
  nudges_lr = c(0, 0),
  values_are_percentages = TRUE,
  decreasing = FALSE,
  replace_names = FALSE,
  sample_info = NULL,
  show_error = TRUE,
  highlight_significant = FALSE,
  plot_new_data = FALSE,
  labels_data = NULL,
  sec_label = NULL
)

Arguments

model	A model generated with make_xgb_models and has appended predictions with add_predictions.
model_data	The training dataset used to generate the model.
name	The name of the perturbation whose prediction we want to plot.
sample_names	The names of the samples we want to generate a plot for.
n_features	The number of top contributors to show their individual contribution. All other predictors will have their contribution combined.
n_columns	Number of columns to plot in a grid when plotting more than one sample.
short_title	If using many columns, set to TRUE to shorten the title of the plot.
fixed_axis	If TRUE, the plot will be set to a fixed scale (-0.05 to 1). Default = FALSE.
replace_names	If TRUE, the sample name will be replaced using get_cell_line_name. Must supply sample_info.
sample_info	If replace_names is TRUE, this must be supplied.
show_error	If TRUE, a shaded area will be used to visualize prediction interval. Default = TRUE.
highlight_significant	If TRUE, a border will be shown to indicate the prediction interval is above 0.5. Default = FALSE.
plot_new_data	Set to TRUE if plotting the contribution to a new sample (not used in training).

Examples

plot_contribution_to_sample(my_models, model_dataset, "ko_ctnnb1", "my_sample")

Mushriq/mixmap documentation built on Jan. 28, 2024, 7:22 p.m.