scdecon.validation¶

validation ¶

Validation: accuracy metrics for deconvolution against known proportions.

A pure computation layer (NumPy / pandas / scipy.stats) operating on proportion DataFrame objects oriented as cell types x samples. It imports no I/O or plotting.

ValidationReport `dataclass` ¶

ValidationReport(overall_rmse: float, per_type: DataFrame)

Typed, lightweight result of a validation run.

Attributes:

Name	Type	Description
`overall_rmse`	`float`	Root-mean-square error across all cell-type x sample entries.
`per_type`	`DataFrame`	One row per cell type (index), with columns `rmse`, `pearson`, `spearman` (the last two computed across samples; `nan` when a correlation is undefined).

mean_pearson `property` ¶

mean_pearson: float

nan-aware mean of the per-cell-type Pearson correlations.

mean_spearman `property` ¶

mean_spearman: float

nan-aware mean of the per-cell-type Spearman correlations.

to_frame ¶

to_frame() -> pd.DataFrame

Return the per-cell-type metric table (a copy), for serialisation.

render ¶

render() -> str

Return a one-line human-readable summary.

align_proportions ¶

align_proportions(truth: DataFrame, prediction: DataFrame) -> tuple[pd.DataFrame, pd.DataFrame]

Check that truth and prediction describe the same data and align them.

Both must be cell types (index) x samples (columns) with identical cell-type and sample label sets (order-independent). The prediction is reindexed to the truth's cell-type and sample order.

Parameters:

Name	Type	Description	Default
`truth`	`DataFrame`	Ground-truth proportions (cell types x samples).	required
`prediction`	`DataFrame`	Estimated proportions (cell types x samples).	required

Returns:

Type	Description
`tuple[DataFrame, DataFrame]`	`(truth, aligned_prediction)` with matching index and columns.

Raises:

Type	Description
`ValueError`	If either frame is empty, has duplicate cell-type or sample labels, or the cell-type / sample label sets differ.

evaluate ¶

evaluate(truth: DataFrame, prediction: DataFrame) -> ValidationReport

Score estimated proportions against ground truth.