Foldseek on the cluster

Foldseek performs rapid protein structure comparisons and database searches for structural similarity, functional annotation, and clustering. Its CHESS application is named BatchFoldseek.

Prepare directories

Sign in to CHESS. Upload the PDB files to a single cluster input directory:

/hpcfs/fhome/用户名/foldseek/input/

Use a separate output directory:

/hpcfs/fhome/用户名/foldseek/output/

Replace 用户名 with your username. Inputs must be readable; output and temporary directories must be writable.

Submit through CHESS

In Application Center (应用中心), find BatchFoldseek in the biomedical or machine-learning category and add it to your desktop:

Open the application and select Submit (提交):

Field	Meaning
Job name	A short descriptive name
Partition	CPU partition for this task
CPU cores	Adjust to database count and workload
Memory	Requested memory; the form uses MB
Input PDB directory	Cluster folder containing query structures
Output directory	Search results and job scripts
Temporary directory	A writable folder for intermediate files
Databases	One or more target structure databases

The form suggests 4–24 CPU cores and about 3–5 GB RAM per database search. Increase resources as needed for multiple databases and larger inputs. Verify all settings and submit.

Monitor jobs

Open Job Group (作业组) for IDs, states, partitions, CPU allocations, and submission times. A batch may contain multiple subjobs; the number in parentheses indicates the task count.

Demonstration:

Results

Example output listing:

0
1
2
AF-A7MCY6-F1-model_v4-afdb_proteome_aln.tsv
AF-A7MCY6-F1-model_v4-afdb_swissprot_aln.tsv
AF-A7MCY6-F1-model_v4-pdb_aln.tsv
AF-A7MCY6-F1-model_v4.pdb
slurm-276218_0.out
slurm-276218_1.out
slurm-276218_2.out
submit_job.sh

*_aln.tsv files contain alignments and identify the database in their names. slurm-*.out contains subjob logs; submit_job.sh is the generated submission script. Download through CHESS or inspect from the command line:

cd /hpcfs/fhome/用户名/foldseek/output
ls -lh
head AF-A7MCY6-F1-model_v4-pdb_aln.tsv

Troubleshooting

Use full cluster paths, not paths on your local computer. Check that PDB files exist and permissions are correct. If alignments are absent, confirm that all subjobs completed and inspect logs for empty inputs, malformed PDB files, permission failures, or insufficient resources. Searches against each database run separately; increase suitable resources or submit databases in smaller batches if necessary.