Error using PDB and CATH50 downloaded databases

[Fede112]

Expected Behavior

Hi,
I am trying to use the CATH50 and PDB databases downloaded using the foldseek databases module to perform a straightforward search+convertalis with a self-made database containing around 30,000 queries.

Current Behavior

The foldseek convertalis command fails with the following error:
free(): invalid next size (normal)

I managed to run the search by using cath50_seq instead of cath50. I did this because I noticed a problem in the indexing, but I am not sure if this is the correct approach. Is this okay?

Your Environment

Foldseek Version: d326ca0
Also encountered the problem with version: foldseek/8-ef4e960

Thanks in advance

[milot-mirdita]

Could you please post the full terminal output of foldseek?

Proceeding with cath50_seq is fine, this database contains all structures. the cath50 contains only the representative structures, which can be expanded to the cluster members with --cluster-search 1.

But it shouldn't crash in either case.

[Fede112]

Hi,

I conducted further tests, and here are the outputs for the PDB database. They are essentially the same as the CATH50 errors. For comparison, I tested this on a node with Intel processors and a node with ARM processors, and I noticed that the errors are different between the two. I built Foldseek for Intel, so I am not too worried about ARM not working. Strangely enough, both work with pdb_seq.

log_arm_stderr.txt
log_arm_stdout.txt
log_intel_stderr.txt
log_intel_stdout.txt

[milot-mirdita]

Does it work correctly when you turn around the roles of query and target (i.e. make pdb the target db)

[Fede112]

yes, if I turn around the roles everything works just fine for either intel or arm.

[milot-mirdita]

We intended for all of these databases from foldseek databases to be used as target databases. I guess we have a bug when using them on the query side.