Yes, using #C02FC6VFQG1 to reconstruct ORFs and quantify them (please check the output in the

results/summary_tables

, particularly

*.counts.tsv.gz

as this is nicely formatted) is the approach I'd advocate. When it comes to making the trees there's more to think about. You can certainly estimate a de novo phylogeny from the ORF sequences. Since your samples are unlikely to contain a good representation of the sequence space for each protein, however, placing in a reference tree (that was estimated from a balanced selection of sequences) is safer. It will also be faster since in phylogenetic placement, each sequence is placed individually. Resource requirements (cpu) will hence scale linearly with the number of sequences (memory is basically flat), which is not the case with full phylogenetics. The only drawback I see is that with placement you will not resolve the individual sequences' branching when they happen to branch off the reference tree at the same place. In other words, quite similar/closely related sequences will show up as a "multitomy", a brush in the resulting tree.