(we have recovered isolates from fecal samples for species which were not detected by PanPhlAn3 [presumably due to low coverage/abundance]).
Maybe this can be due to parameters tuning of strain detection, in this topic we discussed this kind of issue :
If that still does not work, the last method you mentioned (“Eg. aligning every ORF to the PanPhlAn species database instead”)
In the topic linked above, I explain how the coverage threshold work and the kind of profile PanPhlAn expect from a natural microbiota sample. I’m not sure making a fq file out of the genome will be treatable by PanPhlAn, as house-keeping genes present in other species are expected at very high coverage while parts of the accessory genome should have a low coverage compared to the rest of the species genome.