Estimated reference expression signatures

[wangjiawen2013]

Hi, The samples sequenced in more depth will have more counts. So, are the estimated reference expression signatures generated by mod.export_posterior() normalized or scaled or just the inferred counts?

[vitkl]

Citing supplementary methods page 57, Negative Binomial regression model estimates average expression of each gene in each cell type while accounting for a number of technical factors. The reference signatures are not normalised or scaled in the classical meaning of normalisation (dividing by total count) and scaling (subtracting per gene mean and dividing by standard deviation). Instead, during inference, the model removed the batch effects (see which effects below) from the average expression of each gene in each cell type.