Calculate Network Prediction Performance Metrics

Source: R/network_evaluation.R

Calculates comprehensive performance metrics for evaluating predicted network structures, including classification performance, precision-recall metrics, and network topology metrics.

calculate_metrics(
  network_table,
  ground_truth,
  type = "all",
  plot = FALSE,
  line_color = "#1563cc",
  line_width = 1
)

Arguments

network_table

A data frame of predicted network structure containing:

  • regulator - Source nodes of the network edges

  • target - Target nodes of the network edges

  • weight - Edge weights representing prediction confidence

ground_truth

A data frame of ground truth network with the same format as network_table.

type

The type of metric to return, default is all. This can take any of the following choices:

  • all - Returns all available metrics

  • auroc - Area Under ROC Curve, evaluating overall classification ability

  • auprc - Area Under Precision-Recall Curve, focusing on positive class prediction

  • precision - Proportion of correct predictions among positive predictions

  • recall - Proportion of actual positives correctly identified

  • f1 - Harmonic mean of precision and recall

  • si - Set Intersection, counting correctly predicted edges

  • ji - Jaccard Index, measuring overlap between predicted and true networks

plot

Logical value, default is FALSE, whether to generate visualization plots. If TRUE, generates:

  • ROC curve plot for AUROC evaluation

  • Precision-Recall curve plot for AUPRC evaluation

  • Confusion matrix heatmap for classification results

  • Network comparison plot showing edge overlap

  • Metrics summary bar plot

line_color

The color for plot lines, default is #1563cc.

line_width

The width for plot lines, default is 1.

Value

  • If type="all": A data frame with all metrics

  • If specific type: A single numeric value with the requested metric

  • If plot=TRUE: Displays relevant visualization plots

Examples

data("example_matrix")
data("example_ground_truth")
inferCSN(example_matrix) |>
  calculate_metrics(example_ground_truth, plot = TRUE)
#> ✔ Running for <dense matrix>.
#> ✔ Checking input parameters.
#> ✔ Using L0 sparse regression model.
#> ✔ Using 1 core.
#> ✔ Run done.

#>   AUROC AUPRC   ACC Precision Recall    F1  JI SI
#> 1 0.963 0.495 0.944         1  0.514 0.679 0.5 18