Sequence-to-structure search problem

[linzhang-allan]

I used the following commands to perform a sequence-to-structure search with Foldseek:

foldseek createdb reference reference_db
foldseek easy-search test_dataset.fasta reference_db rdrp_Foldseek_Results_prostT5.tsv tmp \
  --format-output "query,target,alnlen,fident,qaln,taln,evalue,bits" \
  --prostt5-model ~/weights \
  --threads 8 \
  --gpu 1

However, in the output file rdrp_Foldseek_Results_prostT5.tsv I get almost no alignments with an e-value < 0.1(29/44804).
Is this expected behavior, or is there something wrong with the way I’m running the command (or with the parameters I’m using)?

[martin-steinegger]

Hard to say what’s the issue without knowing what the exact setup is.

[linzhang-allan]

Thanks for your reply!

When you say “exact setup”, what exactly do you mean? Apart from the options I showed above, I kept all other parameters at their default values.

foldseek easy-search rdrp_test reference_db rdrp_Foldseek_Results_zl.tsv tmp \
  --format-output "query,target,alnlen,fident,qaln,taln,evalue,bits,lddt,alntmscore" \
  -e 10 \
  --min-ungapped-score 0

With this setup, I can achieve high accuracy.

So I’m wondering: could the issue be that the --prostt5-model pipeline is not predicting the 3Di features accurately for my dataset, or am I misunderstanding how it is intended to be used?