net — network structure

Make or modify network structure. Exposed as bobaT.net.

Network construction and feature selection utilities for bobaT.net module.

This module provides functions for: - Building gene regulatory networks from DIRECT-NET and FIGR data - LASSO-based feature selection for network pruning - Network comparison and analysis

bobaT.make_tf_network.prune(G_orig, prune_sources=True, prune_sinks=True)[source]

Prune graph to remove nodes with no incoming or outgoing edges

Parameters:

  • G_orig (networkx.DiGraph) – NetworkX graph to prune

  • prune_sources (bool, optional) – Remove nodes with no incoming edges, defaults to True

  • prune_sinks (bool, optional) – Remove nodes with no outgoing edges, defaults to True

Returns:

Pruned network

Return type:

networkx.DiGraph

bobaT.make_tf_network.prune_info(G_orig, prune_self_loops=True)[source]

Prune graph to only include edges with evidence in multiple databases

Parameters:

  • G_orig (networkx.DiGraph) – NetworkX graph to prune

  • prune_self_loops (bool, optional) – Whether to prune self-loops in the network (edges where parent node == child node), defaults to True

Returns:

Pruned network

Return type:

networkx.DiGraph

bobaT.make_tf_network.prune_to_chea(G_orig, prune_self_loops=True)[source]

Prune graph to only include edges with evidence in ChEA databases

Parameters:

  • G_orig (networkx.DiGraph) – NetworkX graph to prune

  • prune_self_loops (bool, optional) – Whether to prune self-loops in the network (edges where parent node == child node), defaults to True

Returns:

Pruned network

Return type:

networkx.DiGraph

bobaT.make_tf_network.make_network(tfs, outdir='', do_prune=True, prune_sinks=True, prune_sources=True, do_prune_info=True, prune_self_loops=True, do_prune_to_chea=True, save_unfiltered=False, network_name='network')[source]

Make network from list of tfs and save as csv files with various levels of pruning.

Parameters:

  • tfs (List[str]) – List of transcription factor gene names that will be searched in enrichR databases.

  • outdir (str or None) – Output directory where network csvs will be saved, defaults to “”

  • do_prune (bool, optional) – Prune network to remove nodes with no sinks and/or sources and generate a new network file called <network_name>_pruned.csv, defaults to True

  • prune_sinks (bool, optional) – Whether to prune sink nodes from the network (no outgoing nodes), defaults to True

  • prune_sources (bool, optional) – Whether to prune source nodes the network (no incoming nodes), defaults to True

  • do_prune_info (bool, optional) – Whether to prune the network to edges with evidence in more than one database from enrichR and generate a new network file called <network_name>_high_evidence.csv, defaults to True

  • prune_self_loops (bool, optional) – Whether to prune self-loops in the network (edges where parent node == child node), defaults to True

  • do_prune_to_chea (bool, optional) – Whether to prune the network to only edges with evidence in ChEA databases and generate a new network file called <network_name>_chea.csv, defaults to True

  • save_unfiltered (bool, optional) – Whether to save the unfiltered network before pruning, defaults to False. Note that if all pruning options are False and this is set to False, no network will be saved.

  • network_name (str) – Prefix name of network csv files, defaults to network

Returns:

Network with highest level of pruning

Return type:

NetworkX Graph

bobaT.make_tf_network.save_network(network_file, G, attributes=True, overwrite=True)[source]

Save a NetworkX graph to a CSV file.

Parameters:

  • network_file (str) – Path to output file

  • G (networkx.DiGraph) – Network graph to save

  • attributes (bool, default=True) – Whether to save edge attributes

  • overwrite (bool, default=True) – Whether to overwrite existing file

Returns:

True if saved, False if file exists and overwrite=False

Return type:

bool

bobaT.make_tf_network.add_connections(G, net_df, threshold_var=None, threshold=0, parent_node='TF motif', child_node='Target_gene', add_weight=True, weight='motif score', evidence='direct-net')[source]

Add edges to a NetworkX graph from a dataframe.

Parameters:

  • G (networkx.DiGraph) – Network graph to add edges to

  • net_df (pd.DataFrame) – DataFrame containing edge information

  • threshold_var (str, optional) – Column name to threshold on

  • threshold (float, default=0) – Threshold value for filtering edges

  • parent_node (str, default='TF motif') – Column name for source nodes

  • child_node (str, default='Target_gene') – Column name for target nodes

  • add_weight (bool, default=True) – Whether to add edge weights

  • weight (str, default='motif score') – Column name for edge weights

  • evidence (str, default='direct-net') – Evidence type for edge annotation

Returns:

Updated graph with new edges

Return type:

networkx.DiGraph

bobaT.make_tf_network.build_network_from_directnet(direct_net_df, threshold=0, threshold_var='motif score')[source]

Build a gene regulatory network from DIRECT-NET data.

Parameters:

  • direct_net_df (pd.DataFrame) – DIRECT-NET data with columns: ‘TF motif’, ‘Target_gene’, ‘motif score’

  • threshold (float, default=0) – Minimum motif score threshold

  • threshold_var (str, default='motif score') – Column name to threshold on

Returns:

Gene regulatory network

Return type:

networkx.DiGraph

bobaT.make_tf_network.lasso_feature_selection(network, data, network_name, output_dir, alphas=None, save_network=True, plot=True)[source]

Perform LASSO regression for feature selection on network edges.

For each target gene, use LASSO to identify the most important transcription factor regulators from the network.

Parameters:

  • network (pd.DataFrame) – Network dataframe with columns: ‘source’, ‘target’, ‘weight’, ‘evidence’

  • data (pd.DataFrame) – Expression data with genes as columns

  • network_name (str) – Name for output files

  • output_dir (str) – Directory for saving results

  • alphas (list, optional) – Alpha values to test. Default: [0.00001, 0.0001, 0.001, 0.01, 0.1, 1]

  • save_network (bool, default=True) – Whether to save filtered networks

  • plot (bool, default=True) – Whether to generate coefficient plots

Returns:

Dictionary mapping alpha values to best scores

Return type:

dict

bobaT.make_tf_network.compare_networks(original_network, filtered_network, alpha, save=True, save_dir='')[source]

Compare original network to filtered network and visualize differences.

Parameters:

  • original_network (pd.DataFrame) – Original network with ‘source’ and ‘target’ columns

  • filtered_network (pd.DataFrame) – Filtered network with ‘source’ and ‘target’ columns

  • alpha (float) – Alpha value used for filtering

  • save (bool, default=True) – Whether to save the histogram

  • save_dir (str, default="") – Directory for saving plots

Returns:

Summary statistics about the network comparison

Return type:

dict

bobaT.make_tf_network.adaptive_lasso_pruning(network, output_dir, network_name, max_parents=8, alphas=None)[source]

Adaptively select alpha for each target to achieve desired max parents.

For targets with more than max_parents regulators, find the smallest alpha that reduces regulators to max_parents or fewer.

Parameters:

  • network (pd.DataFrame) – Network dataframe with ‘source’ and ‘target’ columns

  • output_dir (str) – Directory containing LASSO results

  • network_name (str) – Name of the network

  • max_parents (int, default=8) – Maximum number of regulators per target

  • alphas (list, optional) – Alpha values to search. Default: [0.00001, 0.0001, 0.001, 0.01, 0.1, 1]

Returns:

  • pd.DataFrame – Pruned network with at most max_parents regulators per target

  • dict – Dictionary mapping each target to its selected alpha