plot_pseudobeta {Coxmos}

plot_pseudobeta

Description

This function decomposes a PLS-Cox model, translating it into a pseudo-beta interpretation with respect to the original variables. The decomposition is based on the relationship between the Cox coefficients associated with each component and the weights corresponding to the original variables. The final Cox formula is thus expressed in terms of these original variables.

Usage

plot_pseudobeta(
  model,
  error.bar = TRUE,
  onlySig = FALSE,
  alpha = 0.05,
  zero.rm = TRUE,
  top = NULL,
  auto.limits = TRUE,
  show_percentage = TRUE,
  size_percentage = 3,
  title_size_text = 15,
  legend_size_text = 12,
  x_axis_size_text = 10,
  y_axis_size_text = 10,
  label_x_axis_size = 10,
  label_y_axis_size = 10
)

Arguments

model

Coxmos model.

error.bar

Logical. Show error bar (default: TRUE).

onlySig

Logical. Compute pseudobetas using only significant components (default: FALSE).

alpha

Numeric. Numerical values are regarded as significant if they fall below the threshold (default: 0.05).

zero.rm

Logical. Remove variables with a pseudobeta equal to 0 (default: TRUE).

top

Numeric. Show "top" first variables with the higher pseudobetas in absolute value. If top = NULL, all variables are shown (default: NULL).

auto.limits

Logical. If "auto.limits" = TRUE, limits are detected automatically (default: TRUE).

show_percentage

Logical. If show_percentage = TRUE, it shows the contribution percentage for each variable to the full model (default: TRUE).

size_percentage

Numeric. Size of percentage text (default: 3).

title_size_text

Numeric. Text size for legend title (default: 15).

legend_size_text

Numeric. Text size for legend title (default: 12).

x_axis_size_text

Numeric. Text size for x axis (default: 10).

y_axis_size_text

Numeric. Text size for y axis (default: 10).

label_x_axis_size

Numeric. Text size for x label axis (default: 10).

label_y_axis_size

Numeric. Text size for y label axis (default: 10).

Details

The plot_pseudobeta function offers a comprehensive visualization and interpretation of a PLS-Cox model in terms of the original variables. The function begins by validating the model's class and type. For single block models, the function computes the pseudo-betas by multiplying the loading weights (W.star) with the Cox coefficients. For multiblock models, this computation is performed for each block separately.

The function provides flexibility in terms of visualization. Users can opt to display error bars, filter out non-significant components based on a significance threshold (alpha), and remove variables with a pseudo-beta of zero. Additionally, the function allows for automatic limit detection for the plot and displays the contribution percentage of each variable to the full model. The resulting plot can be customized further with various text size parameters for different plot elements.

It's worth noting that the function supports both single block and multiblock PLS-Cox models. For multiblock models, the function returns a list of plots, one for each block, whereas for single block models, a single plot is returned.

Value

A list containing the following elements: plot: Depending on the model type, this can either be a single ggplot object visualizing the pseudo-beta coefficients for the original variables in a single block PLS-Cox model, or a list of ggplot objects for each block in a multiblock PLS-Cox model. Each plot provides a comprehensive visualization of the pseudo-beta coefficients, potentially including error bars, significance filtering, and variable contribution percentages. beta: A matrix or list of matrices (for multiblock models) containing the computed pseudo-beta coefficients for the original variables. These coefficients represent the influence of each original variable on the survival prediction. sd.min: A matrix or list of matrices (for multiblock models) representing the lower bounds of the error bars for the pseudo-beta coefficients. sd.max: A matrix or list of matrices (for multiblock models) representing the upper bounds of the error bars for the pseudo-beta coefficients.

Author(s)

Pedro Salguero Garcia. Maintainer: [e-mail]

Examples

data("X_proteomic")
data("Y_proteomic")
X <- X_proteomic[,1:50]
Y <- Y_proteomic
splsicox.model <- splsicox(X, Y, n.comp = 2, penalty = 0.5, x.center = TRUE, x.scale = TRUE)
plot_pseudobeta(model = splsicox.model)