MetaPhlAn preprocessing of reads

Hello, I have a question about what are the best practices to pre process my metagenomic data ( whole genome) reads properly to best suit the metaphlan tool.
Is paired-end merging (gives longer reads) before hand better ?
And I saw papers where they get rid of the human reads before using metaphlan (with bowtie) and only analyze the unmapped reads. Is this mandatory ?

Thank you

I would avoid merging the paired-end reads and feed them as single end since bowtie2 is optimized with short reads. I would encourage you to remove any kind of host-associated but is not required as the markers will no map against the human reads