Workflow for DE after high-resolution CibersortX deconvolution

Question

[Aspire]

The authors of CibersortX mainly used TPM in their analysis as the unit of gene expression in samples (Supplementary table 1 in article)

The High-Resolution feature in CibersortX outputs deconvolution values per-gene per-sample, in theory making it possible to test for differential expression. However, TPM units are not recommended for differential expression.

So it is "good" to use TPM as input for CibersortX, but "bad" to use it for differential expression following it.

Is using TPM units both as input for CibersortX and for DE following it - best practice? Is there a better alternative?