liana.mt.find_causalnet

Contents

liana.mt.find_causalnet#

liana.mt.find_causalnet(prior_graph, input_node_scores, output_node_scores, node_weights=None, node_cutoff=0.1, min_penalty=0.01, max_penalty=1.0, missing_penalty=10, edge_penalty=0.01, solver=None, seed=1337, max_runs=1, stable_runs=5, verbose=True, **kwargs)#

Find the causal network that best explains the input/output node scores.

Parameters:
prior_graph Graph

The prior graph to use for the search.

input_node_scores Mapping[str, float]

A dictionary of input node scores.

output_node_scores Mapping[str, float]

A dictionary of output node scores.

node_weights Mapping[str, float] | None (default: None)

A dictionary of node weights. The keys are the node names, the values are the weights. If None, all nodes will have the same weight.

node_cutoff float (default: 0.1)

The cutoff to use for the node weights. Nodes with a weight below this cutoff will be assigned the max_penalty, nodes with a weight above this cutoff will be assigned the min_penalty. Only used if node_weights is not None. Default: 0.1

min_penalty float (default: 0.01)

The minimum penalty to assign to nodes with a weight above the cutoff. Only used if node_weights is not None. Default: 0.01

max_penalty float (default: 1.0)

The maximum penalty to assign to nodes with a weight below the cutoff Only used if node_weights is not None. Default: 1.0

missing_penalty float (default: 10)

The penalty to assign to nodes that are not measured. Default: 10

edge_penalty float (default: 0.01)

The penalty to assign to edges. Default: 0.01

solver str | None (default: None)

The solver to use. If None, the default solver will be used. Default: None It will default to the solver included in SCIPY, if no other solver is available.

seed int (default: 1337)

The seed to use for the random number generator. Default: 1337

max_runs int (default: 1)

The maximum number of runs to perform. Consider increasing this value if the solver does not converge. In each run, the noise added to the edge and node penalties is perturbed slightly (iterating over the seed). By default, only 1 run is performed.

stable_runs int (default: 5)

The number of consecutive stable solutions requires to interrupt the iteration over max_runs. Only used if max_runs is not == 1. Default: 5

verbose bool (default: True)

Whether to print progress information. Default: True

**kwargs object

Additional arguments to pass to the solver.

Return type:

tuple[DataFrame | None, ProblemDef]

Returns:

df_allDataFrame | None

DataFrame containing the resulting causal network

PProblemDef

Insantce of the Corneto problem definition

Examples

Takes the pruned graph from liana.rs.build_prior_network() and selects the sub-network whose signs are consistent with the input and output scores. This needs a mixed-integer solver; without one installed, corneto falls back to the solver bundled with SciPy:

>>> import liana as li
>>> ppis = [("CD4", 1, "LCK"), ("LCK", 1, "JUN"), ("LCK", -1, "FOS")]
>>> prior = li.rs.build_prior_network(ppis, input_nodes={"CD4": 1.0}, output_nodes={"JUN": 1.0, "FOS": -1.0})
>>> df, problem = li.mt.find_causalnet(
...     prior, input_node_scores={"CD4": 1.0}, output_node_scores={"JUN": 1.0, "FOS": -1.0}, verbose=False
... )
Contents