does HUMAnN perform translated search on unclassified reads when the input file is SAM ?
When I read the code, I think it doesn’t.
Sorry for the slow reply. I believe you are correct - HUMAnN interprets SAM input as coming from a successful pangenome mapping and proceeds to quantification from those alignment results. When starting from FASTQ input, the reads that fail to map to pangenomes are isolated during the alignment process (at the same time the SAM file is being built).