Hi, thanks for developing Foldseek!
I wanted to ask something:
I am trying to perform structural searches with Foldseek (specifically using --alignment-type 1 and --exhaustive-search 1) like this:
foldseek easy-search pdbs cath_db results.m8 tmp -c 0.8 --alignment-type 1 --exhaustive-search 1 --exact-tmscore 1 --gpu 1
I am searching on FS's cath database, which I think contains around 800K domains. I am interested in finding hits which have a TM-score>0.5 and min(qcov,tcov)>0.8.
The thing is that when trying to seach the same pdbs on a subset of this database (700 domains) I find that hits with high structural similarity (TM-score > 0.5) found in the small database disappear when searching against the much larger FS-cath database.
The result reporting appears to be gated by the -e parameter, which at the same time seems to depend on the database size. I wanted to know if such E-value is based on 3Di/AA composition, as well as the base size, and if so if there's any way I could make it depend solely on structure and not aminoacid composition. I know that when using --alignment-type 1 the reported E value in the results.m8 file is the meanTMalign(query,target), but increasing -e still affects the reported hits.
I thought of setting -e inf to avoid this gating but the resulting file would be too large when searching in the entire FS-cath database. I tried to reduce its size using something like --tmscore-threshold 0.4 but for some reason the hits don't show up in the results even if the TMalign (mean query and target) was >0.5.
My question would be if I can use Foldseek to report all hits with TM-score>0.5, disregarding any AA composition effects, and regardless of the db size (so like setting -e inf) but at the same time avoid reporting in results.m8 all entries which have TM-score<0.5
Thanks!
Hi, thanks for developing Foldseek!
I wanted to ask something:
I am trying to perform structural searches with Foldseek (specifically using --alignment-type 1 and --exhaustive-search 1) like this:
foldseek easy-search pdbs cath_db results.m8 tmp -c 0.8 --alignment-type 1 --exhaustive-search 1 --exact-tmscore 1 --gpu 1I am searching on FS's cath database, which I think contains around 800K domains. I am interested in finding hits which have a TM-score>0.5 and min(qcov,tcov)>0.8.
The thing is that when trying to seach the same pdbs on a subset of this database (700 domains) I find that hits with high structural similarity (TM-score > 0.5) found in the small database disappear when searching against the much larger FS-cath database.
The result reporting appears to be gated by the -e parameter, which at the same time seems to depend on the database size. I wanted to know if such E-value is based on 3Di/AA composition, as well as the base size, and if so if there's any way I could make it depend solely on structure and not aminoacid composition. I know that when using --alignment-type 1 the reported E value in the results.m8 file is the meanTMalign(query,target), but increasing -e still affects the reported hits.
I thought of setting -e inf to avoid this gating but the resulting file would be too large when searching in the entire FS-cath database. I tried to reduce its size using something like --tmscore-threshold 0.4 but for some reason the hits don't show up in the results even if the TMalign (mean query and target) was >0.5.
My question would be if I can use Foldseek to report all hits with TM-score>0.5, disregarding any AA composition effects, and regardless of the db size (so like setting -e inf) but at the same time avoid reporting in results.m8 all entries which have TM-score<0.5
Thanks!