pylimma.vooma_lm_fit

pylimma.vooma_lm_fit(y, design=None, prior_weights=None, block=None, sample_weights=False, var_design=None, var_group=None, prior_n=10, predictor=None, span=None, legacy_span=False, plot=False, save_plot=False, keep_elist=True, *, key='pylimma', voom_key='vooma', weights_layer='vooma_weights', layer=None)[source]

Combined vooma + lmFit with iterative refinement.

Applies vooma-style mean-variance modelling to expression data, then fits a linear model. If block correlation or sample weights are requested, performs one iteration to refine the estimates.

Parameters:

  • y (ndarray) – Expression matrix (log-scale), shape (n_genes, n_samples).

  • design (ndarray, optional) – Design matrix. If None, uses an intercept-only model.

  • prior_weights (ndarray, optional) – Prior observation weights. Cannot be combined with sample_weights.

  • block (ndarray, optional) – Block factor for correlated samples.

  • sample_weights (bool, default False) – If True, estimate sample-specific quality weights.

  • var_design (ndarray, optional) – Design matrix for variance model (used with sample_weights).

  • var_group (ndarray, optional) – Factor defining variance groups (used with sample_weights).

  • prior_n (float, default 10) – Prior sample size for array weights estimation.

  • span (float, optional) – LOWESS span. If None, chosen adaptively.

  • legacy_span (bool, default False) – If True, use legacy span selection algorithm.

  • save_plot (bool, default False) – If True, include trend data in output.

  • keep_elist (bool, default True) – If True, include expression data with weights in output (dict return path only; AnnData input always writes the weights layer regardless).

  • key (str, default "pylimma") – AnnData adata.uns key for the fit slots (mirrors lm_fit()).

  • voom_key (str, default "vooma") – AnnData adata.uns key for ancillary metadata (span, sample_weights, voom_xy, voom_line). Keeps the fit and the voom-like ancillaries in separate uns buckets, matching the vooma() + lm_fit() split.

  • weights_layer (str, default "vooma_weights") – AnnData output layer for the computed observation weights.

  • layer (str, optional) – AnnData input layer to read expression from. Defaults to adata.X.

  • predictor (ndarray | None)

  • plot (bool)

Returns:

For AnnData input: mutates adata (weights layer, fit in adata.uns[key], ancillaries in adata.uns[voom_key]) and returns None. For ndarray / EList / dict input: returns a dict with the fit plus span, targets (if requested), EList (if keep_elist), and voom_xy / voom_line (if save_plot).

Return type:

dict or None

Notes

This function combines vooma() and lm_fit() with optional iterative refinement of sample weights and intra-block correlation.