From @lfdelzam:

I saw the PathoFact2 subworkflow diagram and it looks good, but it's important to remember that the most important aspect of PathoFact2, more than predicting Virulence Factors and Toxin-related proteins (not just Toxins), is the ability to see everything in context. This means including BGCs (biosynthetic Gene Clusters), MGEs (Mobile Genetic Elements), signal peptides, and ARGs (Antimicrobial Resistance Genes) alongside VF and TOX predictions. Since you're putting everything in a .gff file, these factors should also be included. Therefore, a gene can be VF, TOX, ARG, part of BGC, or present on a plasmid or with signal peptide, or in combinations (VF-TOX, VF-ARG, TOX-BGC).
Here is a file, Genomes_for_PathoFact2_Test_Nextflow.tar 1.gz, containing genomes we used to test PathoFact2; some are known pathogens, others are known non-pathogens. Here you can see the importance of putting everything in context, since many non-pathogens exhibit viral pathogens or toxins (which is known). Therefore, a simple prediction without the genomic context is not very informative.
It would be interesting to see the output you get with these genomes so we can compare it with our output. Does that sound good?

The output of this subworkflow is a GFF that will be integrated with the MGE prediction output in the mobilome annotation pipeline (MAP). I believe we have signal P prediction with InterProscan, but if not, I can add the module to the MAP. We handle BGC prediction in MGnify in another pipeline called mettannotator, but I still have a pending task to create a BGC annotator subworkflow that generates a consensus annotation from three predictors: Gecco, Antismash, and Sanntis (the latter developed within the group). As for ARG, the subworkflow is ready; it just needs to be added to the MAP.

Once I have the subwfs set up in the MAP, I'll generate the outputs with the genomes you shared and send you the results for comparison.