`cca_pvalue_test`¶

Permutation test for the CCA correlation. Shuffles the trajectory labels num_simulations times, recomputes CCA on the first two columns of the named DR, and builds an empirical null distribution of |corr| (the canonical correlation is sign-invariant). Returns the p-value (fraction of permutations whose correlation is at least the observed one) and writes a histogram of the null with the observed value marked.

Source: sample_trajectory/CCA_test.py:519

Signature¶

def cca_pvalue_test(
    pseudo_adata: AnnData,
    column: str,
    input_correlation: float,
    output_directory: str,
    num_simulations: int = 1000,
    trajectory_col: str = "sev.level",
    verbose: bool = True,
) -> float

Parameters¶

Name	Type	Default	Description
`pseudo_adata`	AnnData	—	Sample-level AnnData whose `.obs` rows are samples and whose `.uns` holds the DR coordinates.
`column`	str	—	DR key in `.uns` (the single sample embedding, `"X_DR_sample"`); the first two columns are used.
`input_correlation`	float	—	Observed CCA correlation to test against (from `CCA_Call`).
`output_directory`	str	—	Writes to `{output_directory}/CCA_test/`.
`num_simulations`	int	`1000`	Number of label permutations for the null distribution.
`trajectory_col`	str	`"sev.level"`	Phenotype column in `pseudo_adata.obs`.
`verbose`	bool	`True`	Print timing.

Returns¶

float — empirical p-value.

Output files¶

Under {output_directory}/CCA_test/:

cca_pvalue_distribution_{column}.png
cca_pvalue_result_{column}.txt

Usage¶

from sampledisco.sample_trajectory.CCA_test import cca_pvalue_test

pvalue = cca_pvalue_test(
    pseudo_adata=pseudo_adata,
    column="X_DR_sample",
    input_correlation=cca_score_a,
    output_directory="sampledisco_demo_output/rna",
    num_simulations=1000,
    trajectory_col="sev.level",
)

cca_pvalue_test¶