search

Danko-Lab / TED

a fully Bayesian approach to deconvolve tumor microenvironment
60 stars 10 forks source link

does normalized data input work well? #9

Open yingyonghui opened 3 years ago

yingyonghui commented 3 years ago

Thanks for your amazing tool for deconvolution of the bulk data. The manual said that it accepts raw count matrix from the scRNA-seq, I'm just wondering whether normalized expression matrix (in log scale or something else) would also works well? And does this requirement also apply to the bulk data or not?

Many thanks! Yonghui

tinyi commented 3 years ago

Hi Yonghui,

Thank you for your interest in our tool.

BayesPrism does not work with the log transformed data, as it uses multinomial distribution to model the raw count. You may exponentiate the data and convert back to the raw scale.

BayesPrism is highly robust to linear transformations including RPM, RPKM. Linearly transformed data can be directly used as input in BayesPrism (although raw count is always recommended).

Let me know if there is any other question.

Best,

Tinyi