funannotate2 train error

[samuelmontgomery]

Hi,

I am trying to run funannotate2 (previously used funannotate) on a Scedosporium clinical isolate and am having an error when trying to run train that I can't seem to solve

I know BUSCO doesn't have a scedosporium lineage, so i've used Sordariomycetes as the class
It seems to be failing due to the fact it is not present in the augustus_taxonomy config even though augustus shouldn't be used

looking at train.py it appears to run choose_best_augustus_species no matter what, if I comment out the augustus commands to skip, I get the same error with choose_best_busco_species

funannotate2 train -f funannotate2/sced_aura_clean.fasta -s sordariomycetes -o funannotate2 --cpus 4
[May 16 08:23 AM] Python v3.12.10; funannotate2 v25.4.30; gfftk v25.4.17; buscolite v25.4.24
[May 16 08:23 AM] Loading genome assembly and running QC checks
[May 16 08:23 AM] Genome stats:
{
"n_contigs": 11,
"size": 43308415,
"n50": 5340307,
"n90": 2879831,
"l50": 4,
"l90": 8,
"avg_length": 3937129
}
[May 16 08:23 AM] Getting taxonomy information
{
"kingdom": "Fungi",
"phylum": "Ascomycota",
"class": "Sordariomycetes"
}
Traceback (most recent call last):
File "/home/ubuntu/scratch/software/miniforge3/envs/funannotate2/bin/funannotate2", line 8, in
sys.exit(main())
^^^^^^
File "/home/ubuntu/scratch/software/miniforge3/envs/funannotate2/lib/python3.12/site-packages/funannotate2/main.py", line 28, in main
train(args)
File "/home/ubuntu/scratch/software/miniforge3/envs/funannotate2/lib/python3.12/site-packages/funannotate2/train.py", line 96, in train
aug_species = choose_best_augustus_species(taxonomy)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/scratch/software/miniforge3/envs/funannotate2/lib/python3.12/site-packages/funannotate2/utilities.py", line 368, in choose_best_augustus_species
return best_taxonomy(query_tax, augustus_species)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/scratch/software/miniforge3/envs/funannotate2/lib/python3.12/site-packages/funannotate2/utilities.py", line 423, in best_taxonomy
iMax = max(ref_tax.items(), key=lambda x: x[1][0])[1][0]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: max() iterable argument is empty

[samuelmontgomery]

It appears to be related to the levels in best_taxonomy not matching the levels in busco_taxonomy

i.e., the busco_taxonomy has superkingdom, kingdom, phylum and class which is causing issues matching the taxonomy returned from taxonomy.jgi.doe.gov which has kingdom, phylum and class

altering pretty_taxonomy and best_taxonomy to build the reference and query list dynamically gets through this

def pretty_taxonomy(tax_dict, levels):
    return [tax_dict.get(lvl) for lvl in levels]

def best_taxonomy(query, ref):
    levels = [
        "superkingdom", "kingdom", "phylum", "order",
        "class", "family", "genus", "species"
    ]
    query_tax = {k.lower(): v for k, v in query.items()}
    query_list = pretty_taxonomy(query_tax, levels)

    ref_tax = {}
    for k, v in ref.items():
        ref_list = pretty_taxonomy({rk.lower(): rv for rk, rv in v.items()}, levels)

        # Match only non-None entries at matching positions
        res = [q for q, r in zip(query_list, ref_list) if q and r and q == r]
        if res:
            pos = ref_list.index(res[-1])
            ref_tax[k] = (len(res), len(ref_list[pos:]))

    if not ref_tax:
        print("Warning: No matching taxa found in reference.")
        return []

    iMax = max(ref_tax.items(), key=lambda x: x[1][0])[1][0]
    best = {k: v for k, v in ref_tax.items() if v[0] == iMax}
    iMin = min(best.items(), key=lambda x: x[1][1])[1][1]
    keepers = [k for k, v in best.items() if v[1] == iMin]

    if len(keepers) > 1:
        return random.choice(keepers)
    elif len(keepers) == 1:
        return keepers[0]
    else:
        return []

it is now selecting the most appropriate busco lineage (sordariomycetes) but the augustus species seems to randomly choose a species in the same class (which is as good as it gets I think without a good reference genome)

[nextgenusfs]

Just use -s "Scedosporium sp." and it should fetch taxonomy properly and choose a close enough set of BUSCO models. Idea is not to pick the models manually just enter the most specific taxonomy level you have, ie genus is fine. It won't make a ton of difference on which Augustus training set it picks, there isn't a way to really determine which is "best" and it just uses that species as a starting point for de novo training if Augustus.

[samuelmontgomery]

Thanks - it works well enough after editing the scripts above, still running into errors if I try to run as installed

[nextgenusfs]

Thanks for the fix @samuelmontgomery