I'm analyzing metagenomic count data of open reading frames for several metagenomic samples. I am trying to think of an appropriate normalization method for this data, and have read that it is generally not advisable to simply divide counts by gene-length and a library size factor (i.e. using TPM counts for DESeq2/DE analysis). Why is this not advised? It would seem to allow for within and between sample comparisons and I don't see any intuitive drawbacks. Any help would be greatly appreciated, thank you!