sctransform::vst operates under the assumption that gene counts approximately
follow a Negative Binomial dristribution. For UMI-based data that seems to be
the case, however, non-UMI data does not behave in the same way.
In some cases it might be better to to apply a transformation to such data
to make it look like UMI data. This function applies such a transformation function.
Cells in the input matrix are processed independently. For each cell
the non-zero data is transformed to quantile values. Based on the number of genes
detected a smooth function is used to predict the UMI-like counts.
The functions have be trained on various public data sets and come as part of the
package (see umify_data data set in this package).