bioneuralnet.clustering

Classes

`CorrelatedLouvain`(G, B[, Y, k3, k4, weight, ...])	CorrelatedLouvain Class for Community Detection with Correlated Omics Data.
`CorrelatedPageRank`(graph, omics_data, ...[, ...])	PageRank Class for Clustering Nodes Based on Personalized PageRank.
`HybridLouvain`(G, B, Y[, k3, k4, max_iter, ...])	HybridLouvain Class that combines Correlated Louvain and Correlated PageRank for community detection.

class bioneuralnet.clustering.CorrelatedLouvain(G: Graph, B: DataFrame, Y=None, k3: float = 0.2, k4: float = 0.8, weight: str = 'weight', tune: bool = False, gpu: bool = False, seed: int | None = None)[source]

Bases: object

CorrelatedLouvain Class for Community Detection with Correlated Omics Data. .. attribute:: G

NetworkX graph object.

type:

nx.Graph

B

Omics data.

Type:: pd.DataFrame

Y

Phenotype data.

Type:: pd.DataFrame

k3

Weight for Correlated Louvain.

Type:: float

k4

Weight for Correlated Louvain.

Type:: float

weight

Edge weight parameter name.

Type:: str

tune

Flag to enable tuning of parameters

Type:: bool

get_quality() → float[source]

partition_to_adjacency(partition: dict) → list[source]: Convert the partition dictionary into a list of adjacency matrices (DataFrames), where each adjacency matrix represents a cluster with more than 2 nodes.

run(as_dfs: bool = False) → dict | list[source]

Run correlated Louvain clustering.

If as_dfs is True, returns a list of adjacency matrices (DataFrames), where each adjacency matrix represents a cluster with more than 2 nodes. Otherwise, returns the partition dictionary.

If tune is True and as_dfs is False, hyperparameter tuning is performed and the best parameters are returned. If tune is True and as_dfs is True, tuning is performed, and then standard detection is run using the tuned parameters.

run_tuning(num_samples=10)[source]

class bioneuralnet.clustering.CorrelatedPageRank(graph: Graph, omics_data: DataFrame, phenotype_data: DataFrame, alpha: float = 0.9, max_iter: int = 100, tol: float = 1e-06, k: float = 0.5, tune: bool = False, gpu: bool = False, seed: int | None = None)[source]

Bases: object

PageRank Class for Clustering Nodes Based on Personalized PageRank.

This class handles the execution of the Personalized PageRank algorithm and identification of clusters based on sweep cuts.

alpha

Damping factor for PageRank.

Type:: float

max_iter

Maximum number of iterations for PageRank convergence.

Type:: int

tol

Tolerance for convergence.

Type:: float

k

Weighting factor for composite correlation-conductance score.

Type:: float

output_dir

Directory to save outputs.

Type:: str

generate_weighted_personalization(nodes: List[Any]) → Dict[Any, float][source]

Generates a weighted personalization vector for PageRank.

Parameters:: nodes (List[Any]) – List of node identifiers to consider.
Returns:: Personalization vector with weights for each node.
Return type:: Dict[Any, float]

get_quality() → float[source]: Returns the composite score (or correlation) from the latest clustering run.

phen_omics_corr(nodes: List[Any]) → Tuple[float, str][source]

Calculates the Pearson correlation between the PCA of omics data and phenotype.

Parameters:: nodes (List[Any]) – List of node identifiers to include in the calculation.
Returns:: Correlation coefficient and formatted correlation with p-value.
Return type:: Tuple[float, str]

run(seed_nodes: List[Any]) → Dict[str, Any][source]

Executes the correlated PageRank clustering pipeline.

Steps:

Initializing Clustering:
- Receives a list of seed nodes to personalize the PageRank algorithm.
- Prepares the input graph and relevant parameters for clustering.
PageRank Execution:
- Applies the PageRank algorithm with personalization based on the seed nodes.
- Computes node scores and determines cluster memberships.
Result Compilation:
- Compiles clustering results, including cluster sizes and node memberships, into a dictionary.
- Logs the successful completion of the clustering process.

Args:

seed_nodes (List[Any]):

A list of node identifiers used as seed nodes for personalized PageRank.
These nodes influence the clustering process by biasing the algorithm.

Returns: Dict[str, Any]

A dictionary containing the clustering results. Keys may include:

clusters: Lists of nodes grouped into clusters.

scores: PageRank scores for each node.

metadata: Additional metrics or details about the clustering process.

Raises:

ValueError: If the input graph is empty or seed nodes are invalid.

Exception: For any unexpected errors during clustering execution.

Notes:

Seed nodes strongly influence the clustering outcome; select them carefully based on prior knowledge or experimental goals.

The PageRank algorithm requires a well-defined and connected graph to produce meaningful results.

Results are sensitive to the alpha (damping factor) and other hyperparameters.

run_pagerank_clustering(seed_nodes: List[Any]) → Dict[str, Any][source]

Executes the PageRank clustering algorithm.

Parameters:: seed_nodes (List[Any]) – List of seed node identifiers for personalization.
Returns:: Dictionary containing clustering results.
Return type:: Dict[str, Any]

run_tuning(num_samples: int = 10) → Dict[str, Any][source]

sweep_cut(p: Dict[Any, float]) → Tuple[List[Any], int, float, float, float, str][source]

class bioneuralnet.clustering.HybridLouvain(G: Graph, B: DataFrame, Y: DataFrame, k3: float = 0.2, k4: float = 0.8, max_iter: int = 10, weight: str = 'weight', gpu: bool = False, seed: int | None = None, tune: bool | None = False)[source]