Sample code for small analyses with networkx graph

src/pyBiodatafuse/analyzer/analysis.py

@@ -0,0 +1,100 @@
+import networkx as nx
+import pandas as pd
+from generator import generate_networkx_graph
+
+
+def main():
+    df_test = pd.read_pickle("./combined_df.pkl")
+    print(df_test.head())
+    g = generate_networkx_graph(df_test)
+    print("Number of nodes in graph: {}".format(len(g.nodes)))
+    print(len(g.edges))
+
+    max_out_degree = 0  # node associated to most diseases
+    node_max = None
+    for node in g.nodes:
+        if g.out_degree(node) > max_out_degree:
+            max_out_degree = g.out_degree(node)
+            node_max = node
+
+    print(
+        "Node with most diseases associated: {} with {} disease associations known".format(
+            node_max, max_out_degree
+        )
+    )
+
+    # print(list(g.edges(data=True))[:10])
+    # print(g.get_edge_data("VAMP1", "CHRNE"))
+    # print(list(g.edges(data="label", default=1, keys=True))[0])
+
+    # fetch back all labels in the graph, this can help filter by edge type later
+    labels = set()
+    for u, v, k in g.edges(data=True):
+        labels.add(k["label"])
+
+    print("Labels: {}".format(labels))
+
+    # e.g. retrieving back gene - disease association links only:
+    gene_disease = ((u, v) for u, v, d in g.edges(data=True) if d["label"] == "associated_with")
+
+    # how many edges are of type associated_with?
+    print(len(list(gene_disease)))
+
+    # compute an overview of edge types
+    for label_type in labels:
+        subgraph = ((u, v) for u, v, d in g.edges(data=True) if d["label"] == label_type)
+        print("For label type {} there are {} edges.".format(label_type, len(list(subgraph))))
+
+    # extract interaction subgraph
+    ppi_edges = ((u, v) for u, v, d in g.edges(data=True) if d["label"] == "interacts_with")
+
+    ppi_subgraph = nx.DiGraph()
+    nodes = set()
+    for u, v in ppi_edges:
+        ppi_subgraph.add_node(u)
+        ppi_subgraph.add_node(v)
+        nodes.add(u)
+        nodes.add(v)
+        ppi_subgraph.add_edge(u, v)
+
+    communities_generator = nx.community.girvan_newman(ppi_subgraph)
+    top_level_communities = next(communities_generator)
+    print("Top level communities: {}".format(sorted(map(sorted, top_level_communities))))
+
+    gene_disease_edges = (
+        (u, v) for u, v, d in g.edges(data=True) if d["label"] == "associated_with"
+    )
+
+    gene_disease_subgraph = nx.DiGraph()
+    nodes = set()
+    for u, v in gene_disease_edges:
+        gene_disease_subgraph.add_node(u)
+        gene_disease_subgraph.add_node(v)
+        nodes.add(u)
+        nodes.add(v)
+        gene_disease_subgraph.add_edge(u, v)
+
+    print(len(nodes))
+    communities_generator = nx.community.girvan_newman(gene_disease_subgraph.to_undirected())
+
+    top_level_communities = next(communities_generator)
+    print("Top level communities: {}".format(sorted(map(sorted, top_level_communities))))
+
+    # basic link prediction using Jaccard
+    predictions = list(nx.jaccard_coefficient(gene_disease_subgraph.to_undirected()))
+    non_zero_predictions = [
+        (gene, disease, predicted_score)
+        for (gene, disease, predicted_score) in predictions
+        if predicted_score > 0 and predicted_score != 1
+    ]
+
+    # sort by prediction score descending
+    non_zero_predictions.sort(key=lambda x: x[2], reverse=True)
+
+    # print top 10 predictions
+    # ... actually predicts disease - disease links !
+    print("Predicted links: {}".format(non_zero_predictions[:10]))
+
+
+if __name__ == "__main__":
+    main()