INNER CODE UNIT · Python

disp_residuals

scverse/PyDESeq2 · src/pydeseq2/dds.py:829

        disp_residuals = np.log(
            self[:, self.non_zero_genes].var["genewise_dispersions"]
        ) - np.log(self[:, self.non_zero_genes].var["fitted_dispersions"])

        # Compute squared log-residuals and prior variance based on genes whose
        # dispersions are above 100 * min_disp. This is to reproduce DESeq2's behaviour.
        above_min_disp = self[:, self.non_zero_genes].var["genewise_dispersions"] >= (
            100 * self.min_disp
        )

        self.uns["_squared_logres"] = (
            mean_absolute_deviation(disp_residuals[above_min_disp]) ** 2
        )

        self.uns["prior_disp_var"] = np.maximum(
            self.uns["_squared_logres"] - polygamma(1, (num_samples - num_vars) / 2),
            0.25,
        )

View source record →