Creating a NetworkGraph

RISK generates a NetworkGraph object for overrepresentation-based analysis and visualization. This step integrates the network, annotation, and overrepresentation results into a cohesive structure with clustering, domain-level significance, and downstream plotting support.

---

Usage

To build a NetworkGraph, you’ll need:

• A preprocessed network (e.g., from load_network_*())
• An annotation object (from load_annotation_*())
• A neighborhoods object (from one of the overrepresentation tests)

graph = risk.load_graph(
    network=network,
    annotation=annotation,
    neighborhoods=neighborhoods,
    tail="right",
    pval_cutoff=0.05,
    fdr_cutoff=1.00,
    impute_depth=0,
    prune_threshold=0.125,
    linkage_criterion="distance",
    linkage_method="single",
    linkage_metric="jaccard",
    linkage_threshold="auto",
    min_cluster_size=6,
    max_cluster_size=1_000,
)

---

Parameters

Parameter	Description
network	A preprocessed network
annotation	An annotation object
neighborhoods	A neighborhoods object
tail	Test direction: 'right' (default), 'left', or 'both'
pval_cutoff	Raw p-value threshold (e.g., 0.01–0.05)
fdr_cutoff	Adjusted p-value threshold (e.g., 0.05–1.0)
impute_depth	Hop depth for stratified overrepresentation imputation
prune_threshold	Prune distant nodes from cluster layout
linkage_criterion	Clustering criterion. Choose from: 'distance', 'maxclust', or 'off' (disable clustering).
linkage_method	Linkage method for clustering. Choose from: 'single', 'complete', 'average', 'weighted', 'centroid', 'median', 'ward', or 'auto' (automatically selects best method).
linkage_metric	Distance metric for clustering. Choose from: 'jaccard', 'cosine', 'euclidean', 'cityblock', 'yule', 'braycurtis', 'hamming', and more; or use 'auto' to select automatically.
linkage_threshold	Float (0–1) or 'auto' for optimization
min_cluster_size	Minimum size for functional domains (clusters)
max_cluster_size	Maximum size for functional domains

---

Key Attributes

After creation, the NetworkGraph object contains cluster- and node-level mappings:

Domain-Level

• domain_id_to_node_ids_map
• domain_id_to_node_labels_map
• domain_id_to_domain_terms_map
• domain_id_to_domain_info_map

Node-Level

• node_id_to_node_label_map
• node_label_to_node_id_map
• node_label_to_significance_map
• node_significance_sums

These mappings drive all visualization, labeling, and export methods.

---

Analysis Summary

You can extract the analysis results using .summary.load():

summary_df = graph.summary.load()
summary_df.head()

Export formats:

graph.summary.to_csv("./data/csv/summary/michaelis_2023.csv")
graph.summary.to_json("./data/json/summary/michaelis_2023.json")
graph.summary.to_txt("./data/txt/summary/michaelis_2023.txt")

---

Cleaning Domains

Use .pop() to remove a domain from all internal structures. This method returns the node labels associated with the removed domain.

domain_1_labels = graph.pop(1)

---

Next Step

Proceed to 6. Visualization to plot overrepresented clusters and explore domain layout.