Metrics

The metrics module includes functions that perform statistical analyses and generate visualizations. Functions from the metrics module allow users to assess and explore network-based embeddings, evaluate clustering performance, and explore the tabular data and network structure.

Correlation Metrics

• bioneuralnet.metrics.omics_correlation() function computes the Pearson correlation coefficient between the first principal component of the omics data and a phenotype. The data are standardized and reduced in dimension by PCA before correlation is computed.

• bioneuralnet.metrics.cluster_correlation() function computes the Pearson correlation for a cluster of features with a phenotype. Clusters with fewer than two features or zero variance are handled appropriately.

Visualization

The module also contains several plotting functions:

• bioneuralnet.metrics.plot_variance_distribution() plots the distribution of feature variances …

• bioneuralnet.metrics.plot_variance_by_feature() shows the variance of each feature against its index.

• bioneuralnet.metrics.plot_performance() compares performance metrics between raw and enhanced omics data.

• bioneuralnet.metrics.plot_embeddings() visualizes high-dimensional embeddings in 2D using t-SNE.

• bioneuralnet.metrics.plot_network() displays the network graph from an adjacency matrix.

• bioneuralnet.metrics.compare_clusters() compares clusters from different methods.

Evaluation

Functions to train and evaluate RandomForest and XGBoost models over one or multiple runs (Mostly used internally for testings purposes):

• bioneuralnet.metrics.evaluate_model() evaluates a model over multiple runs.

• bioneuralnet.metrics.evaluate_single_run() runs a single evaluation loop.

• bioneuralnet.metrics.evaluate_rf() evaluates a Random Forest model.

• bioneuralnet.metrics.evaluate_xgb() evaluates an XGBoost model.

• bioneuralnet.metrics.evaluate_f1w() computes the weighted F1 score.

• bioneuralnet.metrics.evaluate_f1m() computes the macro F1 score.

• bioneuralnet.metrics.compare_clusters() compares and plots clusters from different methods.

Example Usage

1. Variance Distribution The following example generates and plots the distribution of feature variances:

from bioneuralnet.metrics import plot_variance_distribution

fig = plot_variance_distribution(omics2, bins=100)

—

2. Variance Per Feature Plots variance for the first 20 features.

from bioneuralnet.metrics import plot_variance_by_feature

fig2 = plot_variance_by_feature(omics2.iloc[:, 0:20])

3. GNN Embeddings Training a GNN-based embedding model and visualizing the embeddings.

from bioneuralnet.network_embedding import GNNEmbedding

merged_omics = pd.concat([omics1, omics2], axis=1)

gnn = GNNEmbedding(
    adjacency_matrix=global_network,
    omics_data=merged_omics,
    phenotype_data=phenotype,
    clinical_data=clinical,
    phenotype_col="phenotype",
)

gnn.fit()
embeddings = gnn.embed(as_df=True)
display(embeddings.head())

from bioneuralnet.metrics import plot_embeddings

# Using our embeddings instance, we get the necessary labels for the graph.
node_labels = gnn._prepare_node_labels()
embeddings_array = embeddings.values

embeddings_plot = plot_embeddings(embeddings_array, node_labels)

4. Network Visualization This section visualizes Louvain clusters in a network format.

from bioneuralnet.metrics import plot_network
from bioneuralnet.metrics import louvain_to_adjacency

cluster1 = louvain_clusters[0]
cluster2 = louvain_clusters[1]

# Convert Louvain clusters into adjacency matrices
louvain_adj1 = louvain_to_adjacency(cluster1)
louvain_adj2 = louvain_to_adjacency(cluster2)

# Plot using the converted adjacency matrices
cluster1_mapping = plot_network(louvain_adj1, weight_threshold=0.12, show_labels=True, show_edge_weights=False)
display(cluster1_mapping.head())

cluster2_mapping = plot_network(louvain_adj2, weight_threshold=0.12, show_labels=True, show_edge_weights=False)
display(cluster2_mapping.head())

Further Information

For more details on each function and its parameters, please refer to the inline documentation in the source code. Our GitHub repository is available from the index page.