Hi,
I ran HUMAnN 4.0.0 alpha1 using both --count-normalization "Counts" and --count-normalization "RPKs" on the same dataset.
I compared the corresponding *2_genefamilies.tsv output files and expected the relationship RPKs = Counts / gene length (in kb).
To test this, I retrieved gene lengths from the *_bowtie2_aligned.tsv or *_diamond_aligned.tsv intermediate files, depending on whether the gene family was detected through the nucleotide or translated search. However, the calculated values did not match the reported RPKs.
I understand that Counts are not necessarily integer read counts, since reads can be assigned fractionally across multiple targets, but could someone explain how RPKs are calculated in HUMAnN 4.0.0 alpha1 and how they relate to Counts?
Many thanks!